DREKF-based real-time online diagnosis method for abnormal fault of valve clearance of engine

Abstract

The invention aims to provide a DREKF-based real-time online diagnosis method for an abnormal fault of an engine valve clearance, and belongs to the field of engine diagnosis. Comprising the following steps: acquiring an engine cylinder cover vibration signal in real time, establishing and simplifying a cylinder cover vibration signal mathematical model, and determining an undetermined parameter; selecting a proper state variable to construct a cylinder cover vibration signal state space equation, and verifying observability of the cylinder cover vibration signal state space equation; designing a dual-rate extended Kalman filtering optimal estimation method, and inputting the vibration signal into the DREKF to obtain a cylinder cover vibration signal estimation value; meanwhile, a process derivative and a derivative threshold value of a vibration signal estimated based on the DREKF are obtained through calculation, and on-line real-time fault diagnosis of the valve clearance of the engine is achieved according to judgment and accurate measurement of the state of the valve clearance. According to the invention, a novel engine cylinder cover vibration signal nonlinear state space model is established, the model is suitable for observation of engine cylinder cover vibration signals under different rotating speeds and different loads, and online real-time diagnosis of the valve clearance state is realized.

Description

Technical Field

[0001] The invention relates to an engine diagnosis method, in particular to an abnormal valve clearance fault diagnosis method. Background Art

[0002] The cylinder head vibration signal of the engine contains a wealth of working status information of various engine components, such as changes in combustion pressure in the cylinder, opening and closing of the valve, friction between the piston and the cylinder liner, etc. Through real-time analysis and processing of the cylinder head vibration signal, it can be used for engine fault diagnosis, combustion state evaluation and other aspects. Abnormal engine valve clearance failure is one of the typical engine failures. If it is not diagnosed and repaired in time, it will lead to reduced engine mechanical efficiency and power, affecting the safe operation of equipment and systems, and even affecting the personal safety of operators. Therefore, conducting research on abnormal valve clearance fault diagnosis is of great significance to ensure the safe and efficient operation of the engine.

[0003] The paper "Diagnosis Method of Abnormal Valve Clearance Fault of Diesel Engine Based on Complex Morlet Transform and Improved AlexNet Neural Network" (Journal of Beijing University of Chemical Technology, 2021) diagnosed the abnormal valve clearance fault of TBD234V12 diesel engine, selected the complex Morlet wavelet basis to perform continuous wavelet transform on the cylinder head vibration signal, compressed and normalized the transformed time-frequency graph, and input it into the AlexNet neural network, and improved the AlexNet neural network structure, and found the optimal parameters by adjusting the parameters to make the fault diagnosis model accuracy reach 98.91%. The method used in this paper has high computing resource requirements; the diagnostic method is data-driven and highly data-dependent; the model has poor interpretability, and the neural network is like a "black box", making it difficult to understand the internal decision-making process and feature extraction mechanism. The paper "Application of Improved Variational Mode Decomposition in Diagnosis of Abnormal Valve Clearance Fault of Engine" (Journal of Beijing University of Chemical Technology, 2021) diagnosed the abnormal valve clearance fault of TBD234V12 diesel engine and proposed an improved VMD method, which optimizes the number of modes and penalty factors based on the principle of minimum correlation coefficient of adjacent frequency band signals and maximum entropy of high-frequency noise power spectrum, and verifies its effectiveness through simulation signals. Based on this, a fault diagnosis method was constructed to form a complete diagnostic process. Through valve clearance fault experiments under different working conditions, it was finally verified that this method has a higher recognition rate than the traditional VMD method in fault identification, with an average recognition rate of 86.9%. The VMD method used in this paper takes a long time to calculate, and its universality for parameters of different diesel engines and different working conditions is insufficient. It is also data-driven, and the large demand for computing resources leads to complex calculations and limited real-time performance, which affects the effective application of this method in a wider range of scenarios and real-time monitoring. Summary of the invention

[0004] The object of the present invention is to provide a real-time online diagnosis method for abnormal engine valve clearance fault based on DREKF, which is applicable to observation of engine cylinder head vibration signals at different speeds and different loads.

[0005] The object of the present invention is achieved in that:

[0006] The real-time online diagnosis method of abnormal engine valve clearance fault based on DREKF of the present invention is characterized by comprising the following steps:

[0007] (1) According to the characteristics of the engine cylinder head vibration signal, a nonlinear mathematical model of the engine cylinder head vibration signal is established and simplified;

[0008] (2) Vibration signal parameter identification: The impact frequency and attenuation damping parameters of the target cylinder head vibration signal are identified based on the Laplace wavelet correlation filter algorithm;

[0009] (3) Perform initial parameter identification on the cylinder head vibration signal to select state variables, and construct a dual-rate nonlinear state space model based on the state variables. Discretize the nonlinear state space model and verify the observability of the model.

[0010] (4) The process noise and measurement noise are taken into account in the model, and the model is linearized to obtain the linear discrete system state space equation. Based on the dual-rate extended Kalman filter algorithm, the optimal estimation method of the engine cylinder head vibration signal is designed;

[0011] (5) The basic parameters of the DREKF algorithm are identified and set. Then, the process derivative obtained by estimating the cylinder head vibration signal based on DREKF in step (3) is used as the monitoring target, the specific parameters and thresholds of fault diagnosis are defined, and the judgment criteria of the valve clearance state are set.

[0012] The present invention may also include:

[0013] 1. Step (1) The process of establishing the nonlinear mathematical model of the engine cylinder head vibration signal is as follows:

[0014] The cylinder head vibration signal y is approximated as the sum of multiple Laplace wavelet functions:

[0015]

[0016] In the formula, a i is the amplitude of the vibration signal generated by the excitation; ω i is the impact frequency generated by the excitation; i is the attenuation damping generated by the excitation;

[0017] Extract the seating impact component from the cylinder head vibration signal, ignore other vibration components and interference, and the vibration signal is simplified as follows:

[0018]

[0019] Among them, σ is defined as the attenuation damping coefficient; ω is the impact frequency generated by the valve seating impact.

[0020] 2. Step (3) The process of constructing a dual-rate nonlinear state space model is:

[0021] Select cylinder head vibration signal ae σωt sinωt and valve seating impact frequency ω are used as two state variables, and ae is introduced σωt cosωt, then the three state variables are:

[0022]

[0023] The valve seating impact frequency is approximately constant over a period of time, so The update frequency of the cylinder head vibration signal and the impact frequency are inconsistent. The fast sampling vibration signal x1 is defined to be sampled and updated in real time at Δt, and the slow sampling impact frequency x3 is sampled and updated at time interval Δt. f After FFT update, α is defined as the ratio of fast sampling to slow sampling frequency, that is, the inverse of the time interval, and is an integer:

[0024]

[0025] The relationship between the three state variables is:

[0026]

[0027] The dual-rate nonlinear continuous-time state space equation of the engine cylinder head vibration signal is obtained as follows:

[0028]

[0029] In the formula, y a It is the cylinder head vibration signal measurement output of the system; a&f It is the dual measurement output of cylinder head vibration signal and valve seating impact frequency; t f =j·Δt f (j=1,2,...) is the feedback moment of the seating impact frequency;

[0030] Assuming the sampling step length is Δt, the dual-rate nonlinear continuous-time state space equation of the engine cylinder head vibration signal is discretized. The state variable x is k-1 The derivative at time is approximately:

[0031]

[0032] t k,t k-1 Substituting k and k-1, the above formula becomes:

[0033] x(k)=x(k-1)+Δt·f(x(k-1))

[0034] at this time,

[0035]

[0036] The discrete state space model is expressed as:

[0037]

[0038] in,

[0039] According to the dual-rate nonlinear continuous-time state-space equation of the engine cylinder head vibration signal, the Jacobian matrix A is obtained as:

[0040]

[0041] According to the observability matrix, the formula R o =[C CA CA 2 ] T , the rank R of the observability matrix is ​​obtained by calculation o|a =[C1 C1A C1A 2 ] T =3, R o|a&f =[C2 C2A C2A 2 ] T =3, which proves that the state space equation of the nonlinear continuous system shown in the above equation is observable.

[0042] 3. Step (4) The process of designing the optimal estimation method for the engine cylinder head vibration signal is as follows:

[0043] Considering the process noise w(k) and measurement noise v(k), the formula can be Written as:

[0044]

[0045] Among them, w(k), v1(k) and v2(k) are all zero-mean and independent Gaussian white noise, and the covariance matrices are Q(k), R1(k) and R2(k), that is:

[0046] E(w(k))=0,

[0047] E(v1(k))=0,

[0048] E(v2(k))=0, According to the Kalman filter algorithm, let the estimated value of x(k) be A priori estimate and the posterior estimate First, the pair Linearize and obtain the posterior estimate of the state variable x(k) at the sampling time k-1 exist At w(k-1)0=0, perform Taylor series expansion at x(k)=g(x(k-1),w(k-1)), omit the second-order and higher terms, and obtain the approximate linear expression of the state equation:

[0049]

[0050] Among them, A(k-1) and M(k-1) are x(k) in The Jacobian matrix at , we get:

[0051]

[0052] M(k-1) is a constant:

[0053]

[0054] The optimal estimation method of cylinder head vibration signal based on dual-rate extended Kalman filter includes two updating processes: time update and measurement update. Error covariance matrix P(0) + After obtaining the noise covariance matrix Q, R1, and R2, the calculation is performed according to the following time update process and measurement update process, and the system state and covariance matrix are updated repeatedly to finally obtain the optimal estimation result of the cylinder head vibration signal. The specific steps are as follows.

[0055] Time update phase:

[0056] Using nonlinear models Calculate a priori estimates

[0057]

[0058] Calculate the Jacobian matrices A(k-1) and M(k-1) of the state variable x;

[0059] Calculate the prior estimate covariance matrix P(k) - :

[0060] P(k) - =A(k-1)P(k-1) + A(k-1) T+M(k-1)Q(k-1)M(k-1) T

[0061] Measurement update phase:

[0062] ①When k=1,2,...and k≠k f When the measurement update system only updates the cylinder head vibration signal feedback value y a (k), the posterior estimate of the impact frequency of the third state variable and the corresponding error prior remain unchanged, that is:

[0063]

[0064] P (3,3) (k) + =P (3,3) (k-1) +

[0065] The cylinder head vibration measurement signal is used to measure and update the first two state variables. The corresponding measurement update matrix is ​​C a = [1 0], the measurement covariance matrix is ​​R a , define the state variable x at this stage a =[x1 x2] T , the corresponding prior error is recorded as P a (k) - is P(k) - The upper left 2*2 matrix block;

[0066] Calculate the Kalman filter gain K a (k):

[0067] K a (k) = P a (k) - C a (k) T [C a (k)P a (k) - C a (k) T +R a (k)] -1

[0068] Feedback Correction Estimation System:

[0069]

[0070] Calculate the posterior estimation error covariance matrix P a (k) + :

[0071] P a (k)+ =(IK a (k)C a (k))P a (k) - (IK a (k)C a (k)) T +K a (k)R a (k)K a (k) T

[0072] At this point, the estimated output of the system is:

[0073]

[0074] ②When k=1,2,...and k=k f When the measurement update system updates the cylinder head vibration signal and the seat impact frequency feedback value at the same time, the corresponding prior error is recorded as P f (k) - and P(k) - equal, and the corresponding measured output feedback y f (k), measurement update matrix C f , measurement covariance matrix R f They are:

[0075]

[0076] Calculate the Kalman filter gain K f (k), feedback correction estimation system, calculation of the posterior estimation error covariance matrix P f (k) + :

[0077] K f (k) = P f (k) - C f T [C f P f (k) - C f T +R f (k)] -1

[0078]

[0079] P f (k) + =(IK f (k)C f (k))P f (k)- (IK f (k)C f (k)) T +K f (k)R f (k)K f (k) T At this point, the estimated output of the system is:

[0080]

[0081] From the estimated value The estimated output of the system is circulated to continuously update the system state, Kalman filter gain and covariance matrix to achieve real-time estimation of the engine cylinder head vibration signal.

[0082] 4. The process of step (5) is:

[0083] choose As the target value for fault diagnosis monitoring. It can be seen that:

[0084]

[0085] The maximum absolute value of the vibration signal derivative in the healthy state is used as the calculation index, which is defined as:

[0086]

[0087] The threshold of fault diagnosis is defined as the derivative threshold Der th The criteria for judging the healthy status of the engine valve clearance are shown in the following table:

[0088]

[0089] The advantages of the present invention are:

[0090] 1. According to the characteristics of the engine cylinder head vibration signal, the present invention proposes a new mathematical model of the cylinder head vibration signal. On this basis, the cylinder head vibration signal, the vibration signal correlation function and the valve seat impact frequency are used as state variables to establish a new nonlinear state space model of the engine cylinder head vibration signal.

[0091] 2. The present invention discretizes the nonlinear state space model and verifies its observability. In view of the problems of measurement noise and model uncertainty in engineering applications, the measurement noise and process noise are taken into account in the model, the model is linearized, and the model accuracy is improved. The model is suitable for observing the vibration signal of the engine cylinder head at different speeds and different loads.

[0092] 3. In view of the inconsistency between the update frequencies of the cylinder head vibration signal and the valve seating impact frequency, the present invention proposes a real-time estimation method for the cylinder head vibration signal based on a dual-rate extended Kalman filter, which realizes real-time estimation of the engine cylinder head vibration signal through iterative correction of the cylinder head vibration signal and the impact frequency.

[0093] 4. Based on the fault characteristics of engine valve clearance, the present invention proposes a new online real-time diagnosis method for valve clearance status. According to the extended Kalman filter algorithm, a new method for obtaining the process derivative of the vibration signal is proposed, and the fault diagnosis derivative threshold is set according to the process derivative, and the judgment criteria for the abnormal fault diagnosis of the engine valve clearance are set, so as to realize the online real-time diagnosis of the valve clearance status. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 is a flow chart of the present invention;

[0095] Figure 2 It is the normal cylinder head vibration test signal of valve clearance of a single cycle with a speed of 1500rpm and a load of 20kw;

[0096] Figure 3 This is the result of identifying the cylinder head vibration signal under normal valve clearance condition based on Laplace wavelet correlation filtering method at 1500rpm and 20kw.

[0097] Figure 4 The error bar graphs of 10 groups of identification results of cylinder head vibration signals under normal valve clearance conditions based on Laplace wavelet correlation filtering method at 1500rpm and 20kw.

[0098] Figure 5 It is a flow chart of estimating engine cylinder head vibration signal based on DREKF;

[0099] Figure 6 It is the 1s cylinder head vibration test signal with normal valve clearance under the working conditions of 1500rpm and 20kw;

[0100] Figure 7 The cylinder head vibration signal is estimated based on DREKF under the conditions of 1500rpm and 20kw;

[0101] Figure 8 It is a schematic diagram of the process derivative and derivative threshold value obtained by estimating the cylinder head vibration signal with normal valve clearance based on DREKF under the working conditions of 1500rpm and 20kw;

[0102] Fig. 9 It is the 1s cylinder head vibration test signal of abnormal valve clearance under the working conditions of 1500rpm and 20kw;

[0103] Fig.10 It is a schematic diagram of the process derivative of the cylinder head vibration signal with abnormal valve clearance obtained based on DREKF under the working conditions of 1500rpm and 20kw;

[0104] Fig.11 This is a schematic diagram of the process derivative splicing of normal and abnormal valve clearance at 1500rpm and 20kw. DETAILED DESCRIPTION

[0105] The present invention is described in more detail below with reference to the accompanying drawings:

[0106] Example:

[0107] Combination Figure 1-11 The present invention includes: acquiring the engine cylinder head vibration signal in real time, establishing and simplifying the cylinder head vibration signal mathematical model and determining the parameters to be determined; selecting appropriate state variables to construct the cylinder head vibration signal state space equation and verifying its observability; designing a dual-rate extended Kalman filter optimal estimation method, and inputting the vibration signal into DREKF to obtain the cylinder head vibration signal estimation value; at the same time, calculating the process derivative and derivative threshold of the vibration signal based on the DREKF estimation, and accurately determining the valve clearance according to the valve clearance state, so as to realize the online real-time fault diagnosis of the valve clearance of the engine.

[0108] The real-time online diagnosis flow chart of abnormal engine valve clearance fault based on DREKF is as follows: Figure 1 As shown, the following are the main steps of the present invention.

[0109] Step 1: According to the characteristics of the engine cylinder head vibration signal, a nonlinear mathematical model of the engine cylinder head vibration signal is established and simplified.

[0110] Step 2: Identify vibration signal parameters. Based on the Laplace wavelet correlation filter algorithm, the impact frequency, attenuation damping and other parameters of the target cylinder head vibration signal are identified to obtain a more accurate mathematical model and state space equation.

[0111] Step 3: Select appropriate state variables and construct a dual-rate nonlinear state space model based on the state variables. Then discretize the nonlinear state space model and verify the observability of the model.

[0112] Step 4: Take process noise and measurement noise into account in the model, and linearize the model to obtain the linear discrete system state space equation. Then, based on the dual-rate extended Kalman filter algorithm, the optimal estimation method of the engine cylinder head vibration signal is designed.

[0113] Step 5: A real-time online diagnosis method for abnormal valve clearance fault is proposed. The basic parameters of the DREKF algorithm are initially set, and then the process derivative obtained by estimating the cylinder head vibration signal based on DREKF in step 4 is used as the monitoring target characteristic parameter to define the specific parameters and thresholds of fault diagnosis, and set the judgment criteria for the valve clearance state.

[0114] The following is a detailed description:

[0115] Step 1: According to the characteristics of the engine cylinder head vibration signal, a nonlinear mathematical model of the engine cylinder head vibration signal is established and simplified.

[0116] The vibration signal of the engine cylinder head is very complex, mainly including combustion impact, valve seating impact, valve throttling impact and reciprocating inertia force impact of piston connecting rod assembly. Figure 2 This is a single cycle cylinder head vibration signal collected under 1500rpm, 20kw working conditions. According to the characteristics of the vibration signal generated by the impact force acting on the engine cylinder head, the cylinder head vibration signal y can be approximated as the sum of multiple Laplace wavelet functions:

[0117]

[0118] In the formula, a i is the amplitude of the vibration signal generated by the excitation; ω i is the impact frequency generated by the excitation; i is the attenuation damping generated by the excitation.

[0119] Among all the impacts of the cylinder head vibration signal, the valve seating impact is the main component, and when the valve clearance is abnormal, the valve seating impact becomes the component with the most significant change. In order to realize the fault diagnosis of abnormal valve clearance, it is necessary to first extract the seating impact component in the cylinder head vibration signal, ignoring other vibration components and interference, then the vibration signal can be simplified as:

[0120]

[0121] Among them, σ is defined as the attenuation damping coefficient; ω is the impact frequency generated by the valve seating impact.

[0122] Step 2: Based on the Laplace wavelet correlation filtering algorithm, the impact frequency, attenuation damping and other parameters of the target monitored cylinder head vibration signal are identified.

[0123] Due to the similarity between the Laplace wavelet function and the vibration impact signal in the time-frequency domain, the present invention uses the Laplace wavelet correlation filtering method to perform parameter identification. The cylinder head vibration signal and the Laplace wavelet basis function are combined as the inner product to calculate the similarity between the cylinder head vibration signal and the Laplace wavelet at each moment, which is represented by the correlation coefficient k. y :

[0124]

[0125] Where y(t) is the vibration signal; ψ y (t) is the Laplace wavelet function, as shown in formula (2).

[0126] Based on Laplace wavelet correlation filtering algorithm Figure 2 The vibration signal shown in the figure is used for parameter identification, and the identification result is shown in the figure Figure 3 shown.

[0127] Correlation coefficient k y The amplitude and damping of the Laplace wavelet corresponding to the maximum value of is a and σ in equation (2).

[0128] The error bar diagram of the identification results obtained by identifying 10 groups of single-cycle cylinder head vibration signals under 1500rpm and 20kw working conditions based on the Laplace wavelet correlation filtering algorithm is shown in the figure below: Figure 4 shown.

[0129] according to Figure 3 and Figure 4 The identification result diagram shows that the identification result of the cylinder head vibration signal is a=500, ξ=0.14. Calculation shows that σ=-0.14.

[0130] Then, formula (2) can be written as:

[0131] y=500·e -0.14·ωt sinωt (4)

[0132] Step 3: Select appropriate state variables and construct a dual-rate nonlinear state space model based on the state variables. Then discretize the nonlinear state space model and verify the observability of the model.

[0133] When an abnormal valve clearance fault occurs, the cylinder head vibration signal and its valve seat impact frequency will change. σωt sinωt and valve seating impact frequency ω are used as two state variables. In order to more accurately estimate the cylinder head vibration signal, ae is introduced here. σωt cosωt.

[0134] The three state variables are:

[0135]

[0136] Considering that the valve seating impact frequency needs to be obtained based on FFT and is not updated for a period of time, that is, it is approximately constant for a period of time, we have In addition, since the update frequency of the cylinder head vibration signal and the impact frequency are inconsistent, the fast sampling vibration signal x1 is defined to be sampled and updated in real time at Δt, and the slow sampling impact frequency x3 is defined to be sampled and updated at a longer time interval Δt f Updated after FFT.

[0137] The sampling frequency in this embodiment is fs=51200 Hz, so Δt=1 / fs=1.953125×10 -5 s, setting, the impact frequency is updated every 0.2 seconds, then Δt f =0.2s. Here, α is defined as the ratio of fast sampling to slow sampling frequency, that is, the inverse of the time interval, and is an integer:

[0138]

[0139] Then, the relationship between the three state variables is:

[0140]

[0141] The dual-rate nonlinear continuous-time state space equation of the engine cylinder head vibration signal is obtained as follows:

[0142]

[0143] In the formula, y a It is the cylinder head vibration signal measurement output of the system; f It is the dual measurement output of cylinder head vibration signal and valve seating impact frequency; t f =j·Δt f (j=1, 2, ...) is the feedback moment of the seating impact frequency.

[0144] Assuming the sampling step length is Δt, the continuous state space equation of equation (8) is discretized, then the state variable x at t k-1 The derivative at time can be approximated as:

[0145]

[0146] t k ,t k-1 Substituting k and k-1, equation (9) is transformed into:

[0147] x(k)=x(k-1)+Δt·f(x(k-1)) (10)

[0148] at this time,

[0149]

[0150] Then the discrete state space model can be expressed as:

[0151]

[0152] Where, C1=[1 0 0],

[0153] Since formula (8) is a nonlinear model, it is linearized by deriving the Jacobian matrix, and then its observability is judged according to the rank of the observability matrix. According to formula (8), the Jacobian matrix A can be obtained as:

[0154]

[0155] According to the observability matrix, the formula R o =[C CA CA 2 ] T , the rank R of the observability matrix is ​​obtained by calculation o|a =[C1 C1A C1A 2 ] T =3, R o|a&f =[C2 C2A C2A 2 ] T =3, which proves that the state space equation of the nonlinear continuous system shown in equation (8) is observable.

[0156] Step 4: Take process noise and measurement noise into account in the model, and linearize the model to obtain the linear discrete system state space equation. Then, based on the dual-rate extended Kalman filter algorithm, the optimal estimation method of the engine cylinder head vibration signal is designed.

[0157] Considering the process noise w(k) and measurement noise v(k), equation (8) can be written as:

[0158]

[0159] Among them, w(k), v1(k) and v2(k) are all zero-mean and independent Gaussian white noise, and their covariance matrices are Q(k), R1(k) and R2(k) respectively. That is:

[0160]

[0161] According to the Kalman filter algorithm, let the estimated value of x(k) be A priori estimate and the posterior estimate First, linearize equation (14). At sampling time k, according to the Kalman filter algorithm, the posterior estimate of the state variable x(k) at time k-1 can be obtained: exist At w(k-1)0=0, perform Taylor series expansion at x(k)=g(x(k-1),w(k-1)), omit the second-order and higher terms, and obtain the approximate linear expression of the state equation:

[0162]

[0163] Among them, A(k-1) and M(k-1) are x(k) in The Jacobian matrix at , according to formula (14), is obtained as:

[0164]

[0165] M(k-1) is a constant:

[0166]

[0167] The flow chart of estimating engine cylinder head vibration signal based on DREKF is as follows Figure 5 As shown in the figure. The optimal estimation method of cylinder head vibration signal based on DREKF includes two updating processes: time update and measurement update. Error covariance matrix P(0) + After obtaining the noise covariance matrix Q, R1, and R2, the calculation is performed according to the following time update process and measurement update process, and the system state and covariance matrix are continuously updated in a cycle to finally obtain the optimal estimation result of the cylinder head vibration signal.

[0168] The specific steps are as follows:

[0169] Time update phase:

[0170] Calculate the prior estimate using the nonlinear model (14)

[0171]

[0172] Compute the Jacobian matrices A(k-1) and M(k-1) of the state variable x.

[0173] Calculate the prior estimate covariance matrix P(k) - :

[0174] P(k) - =A(k-1)P(k-1) + A(k-1)T +M(k-1)Q(k-1)M(k-1) T (20)

[0175] Measurement update phase:

[0176] ①When k=1,2,...and k≠k f The measurement update system only updates the cylinder head vibration signal feedback value y a (k), the posterior estimate of the impact frequency of the third state variable and the corresponding error prior remain unchanged, that is:

[0177]

[0178] The cylinder head vibration measurement signal is used to measure and update the first two state variables. The corresponding measurement update matrix is ​​C a =

[10] , the measurement covariance matrix is ​​R a . Define the state at this time. The state variable x at this stage a =[x1 x2] T , the corresponding prior error is recorded as P a (k) - is P(k) - The upper left 2*2 matrix block.

[0179] Calculate the Kalman filter gain K a (k):

[0180] K a (k) = P a (k) - C a (k) T [C a (k)P a (k) - C a (k) T +R a (k)] -1 (twenty two)

[0181] Feedback Correction Estimation System:

[0182]

[0183] Calculate the posterior estimation error covariance matrix P a (k) + :

[0184] P a (k) + =(IK a (k)C a (k))P a (k)- (IK a (k)C a (k)) T +K a (k)R a (k)K a (k) T (twenty four)

[0185] At this point, the estimated output of the system is:

[0186]

[0187] ②When k=1,2,...and k=k f When , the measurement update system updates the cylinder head vibration signal and the seat impact frequency feedback value at the same time. The corresponding prior error is recorded as P f (k) - and P(k) - Equal. And the corresponding measurement output feedback y f (k), measurement update matrix C f , measurement covariance matrix R f They are:

[0188]

[0189] Calculate the Kalman filter gain K f (k), feedback correction estimation system, calculation of the posterior estimation error covariance matrix P f (k) + :

[0190] K f (k) = P f (k) - C f T [C f P f (k) - C f T +R f (k)] -1 (27)

[0191]

[0192] P f (k) + =(IK f (k)C f (k))P f (k) - (IK f (k)C f (k)) T +Kf (k)R f (k)K f (k) T (29) At this time, the estimated output of the system is:

[0193]

[0194] According to equations (19) to (30), the system state, Kalman filter gain and covariance matrix are continuously updated to achieve real-time estimation of the engine cylinder head vibration signal.

[0195] Step 5: Initially set the basic parameters of the DREKF algorithm, and then use the process derivative obtained by estimating the cylinder head vibration signal based on DREKF in step 4 as the monitoring target characteristic parameter, define the specific parameters and thresholds of fault diagnosis, and set the judgment criteria for the valve clearance state.

[0196] Set the initial value of the state variable Initial value of error covariance P(0) + , and the initial values ​​of Q, R1, and R2. The selection of Q, R1, and R2 mainly considers the influence of process noise and measurement noise on the filtering effect, and can be selected according to the actual situation. Figure 4 The identification results show that the seating impact amplitude and frequency are 500 and 5710.3 respectively. Therefore, the present invention is set as follows:

[0197]

[0198] In order to verify the estimation effect of the DREKF proposed in the present invention on the cylinder head vibration signal, the cylinder head vibration signal of one second under the condition of normal valve clearance, 1500rpm, 20kw is selected, such as Figure 6 As shown. Figure 6 The cylinder head vibration signal samples shown in the figure are subjected to real-time optimal estimation to obtain the cylinder head vibration signal estimation value as shown in the figure. Figure 7 shown.

[0199] from Figure 6 , Figure 7 By comparison, it can be seen that the DREKF designed based on the present invention can accurately estimate the cylinder head vibration signal, which proves the accuracy of the cylinder head vibration signal state space equation proposed in this paper and the effectiveness of using DREKF to estimate the cylinder head vibration signal.

[0200] The definition of characteristic parameters and thresholds in step 5 and the criteria for determining valve clearance status are described in detail below:

[0201] When an abnormal valve clearance fault occurs in the engine, it is directly manifested as a change in the rate of change of the cylinder head vibration signal, that is, the derivative of the cylinder head vibration signal changes. However, the cylinder head vibration signal contains the influence of measurement noise, and directly taking the derivative of the vibration signal will amplify the noise. Since y = x1, we choose As the target value for fault diagnosis and monitoring. From formula (7), we can know:

[0202]

[0203] The maximum absolute value of the derivative of the vibration signal process under normal valve clearance is used as the calculation index, which is defined here as:

[0204]

[0205] The threshold of fault diagnosis is defined as the derivative threshold Der th As shown in formula (34), the engine valve clearance state judgment criteria are shown in Table 1.

[0206] |Der th |=110%·Der max (34) Table 1 Criteria for judging the status of engine valve clearance

[0207]

[0208] The DREKF designed by the present invention is used for Figure 6 The cylinder head vibration signal samples are used for real-time optimal estimation, and the process derivative and derivative threshold Der of the cylinder head vibration signal with normal valve clearance can be obtained. th like Figure 8 As shown. The maximum absolute value of the process derivative is calculated to be 1.5247*10 6 , the derivative threshold is 1.6771*10 6 Then the valve clearance state judgment criteria set in Table 1 are shown in Table 2.

[0209] Table 2 Criteria for judging the status of engine valve clearance

[0210]

[0211] In order to verify the effectiveness of the valve clearance state determination criterion proposed in the present invention, the cylinder head vibration signal with abnormal valve clearance under the same working conditions of 1500rpm and 20kw is taken as a sample to perform DREKF estimation and real-time monitoring.

[0212] The cylinder head vibration signal sample with abnormal valve clearance under 1500rpm and 20kw conditions is as follows Fig. 9As shown. Applying the DREKF designed in this patent to perform real-time optimal estimation, the process derivative of valve clearance abnormality is obtained as follows Fig.10 shown.

[0213] The process derivatives and derivative thresholds obtained by DREKF for cylinder head vibration signal samples with normal and abnormal valve clearance are spliced ​​together to obtain a comparison chart of process derivatives of normal and abnormal valve clearance, as shown in the figure. Fig.11 shown.

[0214] Fig.11 In the figure, the solid line represents the process derivatives of normal and abnormal valve clearance, and the dashed line represents the derivative threshold calculated under normal valve clearance. The first second of the process derivative is the process derivative of normal valve clearance calculated based on DREKF, and the last second is the process derivative of abnormal valve clearance calculated based on DREKF. Fig.11 It can be seen that the process derivative under the abnormal valve clearance state obviously exceeds the derivative threshold range, which proves the effectiveness of the valve clearance state judgment criterion set in the present invention.

Claims

1. A real-time online diagnosis method for abnormal engine valve clearance fault based on DREKF, characterized by: The steps include: (1) According to the characteristics of the engine cylinder head vibration signal, a nonlinear mathematical model of the engine cylinder head vibration signal is established and simplified; (2) Vibration signal parameter identification: The impact frequency and attenuation damping parameters of the target cylinder head vibration signal are identified based on the Laplace wavelet correlation filter algorithm; (3) Perform initial parameter identification on the cylinder head vibration signal to select state variables, and construct a dual-rate nonlinear state space model based on the state variables. Discretize the nonlinear state space model and verify the observability of the model. (4) The process noise and measurement noise are taken into account in the model, and the model is linearized to obtain the linear discrete system state space equation. Based on the dual-rate extended Kalman filter algorithm, the optimal estimation method of the engine cylinder head vibration signal is designed; (5) The basic parameters of the DREKF algorithm are identified and set. Then, the process derivative obtained by estimating the cylinder head vibration signal based on DREKF in step (3) is used as the monitoring target, the specific parameters and thresholds of fault diagnosis are defined, and the judgment criteria of the valve clearance state are set.

2. The real-time online diagnosis method for abnormal engine valve clearance fault based on DREKF according to claim 1 is characterized by: Step (1) The process of establishing the nonlinear mathematical model of the engine cylinder head vibration signal is as follows: The cylinder head vibration signal y is approximated as the sum of multiple Laplace wavelet functions: In the formula, a i is the amplitude of the vibration signal generated by the excitation; ω i is the impact frequency generated by the excitation; i is the attenuation damping generated by the excitation; Extract the seating impact component from the cylinder head vibration signal, ignore other vibration components and interference, and the vibration signal is simplified as follows: Among them, σ is defined as the attenuation damping coefficient; ω is the impact frequency generated by the valve seating impact.

3. The real-time online diagnosis method for abnormal engine valve clearance fault based on DREKF according to claim 1 is characterized by: The process of constructing the dual-rate nonlinear state space model in step (3) is as follows: Select cylinder head vibration signal ae σωt sinωt and valve seating impact frequency ω are used as two state variables, and ae is introduced σωt cosωt, then the three state variables are: The valve seating impact frequency is approximately constant over a period of time, so The update frequency of the cylinder head vibration signal and the impact frequency are inconsistent. The fast sampling vibration signal x1 is defined to be sampled and updated in real time at Δt, and the slow sampling impact frequency x3 is sampled and updated at time interval Δt. f After FFT update, α is defined as the ratio of fast sampling to slow sampling frequency, that is, the inverse of the time interval, and is an integer: The relationship between the three state variables is: The dual-rate nonlinear continuous-time state space equation of the engine cylinder head vibration signal is obtained as follows: In the formula, y a It is the cylinder head vibration signal measurement output of the system; a&f It is the dual measurement output of cylinder head vibration signal and valve seating impact frequency; t f =j·Δt f (j=1,2,...) is the feedback moment of the seating impact frequency; Assuming the sampling step length is Δt, the dual-rate nonlinear continuous-time state space equation of the engine cylinder head vibration signal is discretized. The state variable x is k-1 The derivative at time is approximately: t k ,t k-1 Substituting k and k-1, the above formula becomes: x(k)=x(k-1)+Δt·f(x(k-1)) at this time, The discrete state space model is expressed as: in, According to the dual-rate nonlinear continuous-time state-space equation of the engine cylinder head vibration signal, the Jacobian matrix A is obtained as: According to the observability matrix, the formula R o =[C CA CA 2 ] T , the rank R of the observability matrix is ​​obtained by calculation o|a =[C1C1A C1A 2 ] T =3, R o|a&f =[C2 C2A C2A 2 ] T =3, which proves that the state space equation of the nonlinear continuous system shown in the above equation is observable.

4. The real-time online diagnosis method for abnormal engine valve clearance fault based on DREKF according to claim 3 is characterized by: Step (4) The process of designing the optimal estimation method for the engine cylinder head vibration signal is as follows: Considering the process noise w(k) and measurement noise v(k), the formula can be Written as: Among them, w(k), v1(k) and v2(k) are all zero-mean and independent Gaussian white noise, and the covariance matrices are Q(k), R1(k) and R2(k), that is: According to the Kalman filter algorithm, let the estimated value of x(k) be A priori estimate and the posterior estimate First, the pair Linearize and obtain the posterior estimate of the state variable x(k) at the sampling time k-1 exist At w(k-1)0=0, perform Taylor series expansion at x(k)=g(x(k-1),w(k-1)), omit the second-order and higher terms, and obtain the approximate linear expression of the state equation: Among them, A(k-1) and M(k-1) are x(k) in The Jacobian matrix at , we get: M(k-1) is a constant: The optimal estimation method of cylinder head vibration signal based on dual-rate extended Kalman filter includes two updating processes: time update and measurement update. Error covariance matrix P(0) + After obtaining the noise covariance matrix Q, R1, and R2, the following time update process and measurement update process are used for calculation, and the system state and covariance matrix are updated repeatedly to finally obtain the optimal estimation result of the cylinder head vibration signal. The specific steps are as follows: Time update phase: Using nonlinear models Calculate a priori estimates Calculate the Jacobian matrices A(k-1) and M(k-1) of the state variable x; Calculate the prior estimate covariance matrix P(k) - : P(k) - =A(k-1)P(k-1) + A(k-1) T +M(k-1)Q(k-1)M(k-1) T Measurement update phase: ①When k=1,2,...and k≠k f When the measurement update system only updates the cylinder head vibration signal feedback value y a (k), the posterior estimate of the impact frequency of the third state variable and the corresponding error prior remain unchanged, that is: P (3,3) (k) + =P (3,3) (k-1) + The cylinder head vibration measurement signal is used to measure and update the first two state variables. The corresponding measurement update matrix is ​​C a =[10], the measurement covariance matrix is ​​R a , define the state variable x at this stage a =[x1 x2] T , the corresponding prior error is recorded as P a (k) - is P(k) - The upper left 2*2 matrix block; Calculate the Kalman filter gain K a (k): K a (k)=P a (k) - C a (k) T [C a (k)P a (k) - C a (k) T +R a (k)] -1 Feedback Correction Estimation System: Calculate the posterior estimation error covariance matrix P a (k) + : P a (k) + =(I-K a (k)C a (k))P a (k) - (I-K a (k)C a (k)) T +K a (k)R a (k)K a (k) T At this point, the estimated output of the system is: ②When k=1,2,...and k=k f When the measurement update system updates the cylinder head vibration signal and the seat impact frequency feedback value at the same time, the corresponding prior error is recorded as P f (k) - and P(k) - equal, and the corresponding measured output feedback y f (k), measurement update matrix C f , measurement covariance matrix R f They are: Calculate the Kalman filter gain K f (k), feedback correction estimation system, calculation of the posterior estimation error covariance matrix P f (k) + : K f (k)=P f (k) - C f T [C f P f (k) - C f T +R f (k)] -1 P f (k) + =(IK f (k)C f (k))P f (k) - (IK f (k)C f (k)) T +K f (k)R f (k)K f (k) T At this point, the estimated output of the system is: From the estimated value The estimated output of the system is circulated to continuously update the system state, Kalman filter gain and covariance matrix to achieve real-time estimation of the engine cylinder head vibration signal.

5. The real-time online diagnosis method for abnormal engine valve clearance fault based on DREKF according to claim 2 is characterized by: The process of step (5) is: choose As the target value for fault diagnosis monitoring, It can be seen that: The maximum absolute value of the vibration signal derivative in the healthy state is used as the calculation index, which is defined as: The threshold of fault diagnosis is defined as the derivative threshold Der th The criteria for judging the healthy status of the engine valve clearance are shown in the following table: 。

Images