CN111783243A - Metal structure fatigue crack propagation life prediction method based on filtering algorithm - Google Patents

Metal structure fatigue crack propagation life prediction method based on filtering algorithm Download PDF

Info

Publication number
CN111783243A
CN111783243A CN202010557546.2A CN202010557546A CN111783243A CN 111783243 A CN111783243 A CN 111783243A CN 202010557546 A CN202010557546 A CN 202010557546A CN 111783243 A CN111783243 A CN 111783243A
Authority
CN
China
Prior art keywords
parameter
state
crack propagation
equation
formula
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010557546.2A
Other languages
Chinese (zh)
Other versions
CN111783243B (en
Inventor
靳慧
陈国良
丁克勤
王立彬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
Original Assignee
Southeast University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Southeast University filed Critical Southeast University
Priority to CN202010557546.2A priority Critical patent/CN111783243B/en
Publication of CN111783243A publication Critical patent/CN111783243A/en
Application granted granted Critical
Publication of CN111783243B publication Critical patent/CN111783243B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/17—Mechanical parametric or variational design

Abstract

The invention discloses a method for predicting the fatigue crack propagation life of a metal structure based on a filtering algorithm, which comprises the following steps: (1) improving the fatigue crack propagation two-parameter index model to obtain an improved fatigue crack propagation two-parameter index model; (2) constructing a state space evaluation model based on the improved fatigue crack propagation two-parameter index model; (3) calculating a state parameter X of the state space evaluation model constructed in the step (2) by using a filtering algorithmk+1(ii) a (4) Using the state parameters
Figure DDA0002544950300000011
The method considers the distribution conditions of the material parameters α and β and the crack length a in the two-parameter index model, and can well predict the fatigue crack propagation life of the metal structure of the casting crane.

Description

Metal structure fatigue crack propagation life prediction method based on filtering algorithm
Technical Field
The invention relates to a method for predicting the fatigue crack propagation life of a metal structure, in particular to a method for predicting the fatigue crack propagation life of the metal structure based on a filtering algorithm.
Background
Since the work grade of the ladle crane structure is high and the ladle crane is often subjected to impact load in the hoisting process, fatigue failure is the main failure mode of the ladle crane. The fatigue crack is the main form of the fatigue damage of the ladle crane, the fatigue crack service life of the metal structure of the ladle crane is researched, and the method has important significance for the use and maintenance of the metal structure of the ladle crane.
The fatigue crack propagation life prediction method has various modes, and the double-parameter index formula has a simple structural form, so that the method is widely applied to the fatigue crack life prediction of the metal structure at present. Two important material parameters alpha and beta in the two-parameter index formula have great influence on the fatigue crack propagation life curve trend, while different structure forms, materials and slight changes in the structure manufacturing process all have influence on alpha and beta, so the two-parameter index formula material parameters alpha and beta are usually obtained by experiments. But the metal structure of the casting crane has complex structure form and dispersed crack propagation parts, and the single fatigue parameters alpha and beta are used for describing the crack propagation life error is large; although the use safety of the structure can be ensured by dividing a certain safety factor, the crack propagation life prediction mode is still different from the crack propagation real state of the structure when facing one crane or even a plurality of cranes.
Therefore, there is a need to solve the above problems.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a metal structure fatigue crack propagation life prediction method based on a filter algorithm, which considers the distribution condition of material parameters alpha and beta in a two-parameter exponential model.
The technical scheme is as follows: in order to achieve the above purpose, the invention discloses a method for predicting the fatigue crack propagation life of a metal structure based on a filtering algorithm, which comprises the following steps:
(1) improving the fatigue crack propagation two-parameter index model to obtain an improved fatigue crack propagation two-parameter index model;
(2) constructing a state space evaluation model based on the improved fatigue crack propagation two-parameter index model;
(3) calculating a state parameter X of the state space evaluation model constructed in the step (2) by using a filtering algorithmk+1；
(4) Using the state parameters
Figure BDA0002544950280000011
And predicting the residual life of the crack propagation.
Wherein the two-parameter exponential model of fatigue crack propagation in the step (1) describes a fatigue crack propagation rate, and the formula is as follows:
Figure BDA0002544950280000021
wherein α is a material parameter, γ is a coefficient related to loading mode, crack shape and position, Δ σ is a stress amplitude, and a is a crack length, and is obtained from formula (1) when α is a fixed parameter, and β is<At 0, the crack propagation rate is increased continuously along with the growth of the crack length, and the increase is less than eαWhen β>0, the crack growth rate decreased with the crack length and was not suitable for life prediction, and it is understood that β<When the crack propagation rate is high, α parameters need to be changed greatly to achieve the prediction effect, and the prediction effect is not beneficial to practical application, so that a two-parameter exponential model is improved as follows:
Figure BDA0002544950280000022
preferably, the constructing of the state space estimation model in the step (2) specifically includes the following steps:
(2.1) constructing a state space parameter evaluation equation:
setting dN as 1, namely the crack propagation length under single cycle load; converting the improved fatigue crack propagation two-parameter exponential model into a discrete recursion form as follows:
Figure BDA0002544950280000023
because the frequent crack parts are dispersed and the stress amplitude on the crack surface is difficult to measure, the delta sigma is replaced by the stress amplitude obtained by finite element simulation; and because the stress amplitude obtained by simulation has calculation error, the stress amplitude delta sigma obtained by simulation is designed to obey normal distribution
Figure BDA0002544950280000024
To simulate the variance of equivalent response amplitude values with true values,
Figure BDA0002544950280000025
is the equivalent stress amplitude; thus ak+1＝f(akΔ σ) in
Figure BDA0002544950280000026
Is aligned with ak+1Performing first-order Taylor expansion to obtain:
Figure BDA0002544950280000027
let ak+1The process noise of (a) is:
Figure BDA0002544950280000028
in the improved fatigue crack propagation two-parameter exponential model, the partial derivative of the delta sigma is calculated as follows:
Figure BDA0002544950280000031
in that
Figure BDA0002544950280000032
In the case where it is known that,
Figure BDA0002544950280000033
and formula (6) are both definite values, and
Figure BDA0002544950280000034
thus, the
Figure BDA0002544950280000035
σaThe expression of (a) is:
Figure BDA0002544950280000036
the crack length a can be obtained from the formula (1), and when the structural characteristics, environment and working conditions cause the fatigue parameters alpha and beta to change, the crack length a is different; then the state space parameter evaluation equation of the two-parameter exponential model is:
Figure BDA0002544950280000037
in the formula (I), the compound is shown in the specification,
Figure BDA0002544950280000038
white Gaussian noise of α, β, as measured by standard practice fatigue testsα,k、σβ,k、σa,kForming a system state parameter variance matrix Q;
(2.2) constructing a state space observation equation
The state space observation equation obtained from the fatigue crack length obtained by monitoring or detection, the corresponding cycle number and the error existing in the measurement process is as follows:
zk+1＝Hk+1xk+1+vk+1(9)
in the formula, Hk+1Is a measurement matrix, here an identity matrix; x is the number ofk+1Is a measured value, here the crack length ak+1；vk+1To measure errors, and is subject to
Figure BDA0002544950280000039
The state space observation equation is as follows:
Zk+1＝ak+1+vk+1(10)
wherein σr,kMoment of variance of composition observationAnd (5) array R.
Further, the step (3) of calculating the state parameter by using the particle filter includes the following specific steps:
initializing state parameters, inputting initial value X of state parameters0＝[a0、α0、β0]And the corresponding variance Q ═ σa,σα,σβ]；
Sampling a set of particles, collecting the set of particles using a probability distribution according to α, a
Figure BDA0002544950280000041
Generating a set of state function samples using a state space parameter estimation equation
Figure BDA0002544950280000042
k +1 is the number of state function steps,
Figure BDA0002544950280000043
is a normal function matrix;
calculating importance weight according to Z in state observation equationk+1Calculating a weight value:
Figure BDA0002544950280000044
wherein the content of the first and second substances,
Figure BDA0002544950280000045
is composed of
Figure BDA0002544950280000046
In the presence of Zk+1The probability of occurrence;
Figure BDA0002544950280000047
is composed of
Figure BDA0002544950280000048
The probability of occurrence;
Figure BDA0002544950280000049
is Zk+1Under the conditions of occurrence
Figure BDA00025449502800000410
The probability of occurrence;
normalizing the weight:
Figure BDA00025449502800000411
random resampling: weighted value according to importance
Figure BDA00025449502800000412
The sample is re-drawn among the N particles,
Figure BDA00025449502800000413
the large number of the distributed particles is more,
Figure BDA00025449502800000414
smaller direct deletion, last by difference
Figure BDA00025449502800000415
Generating new N particles by specific gravity; and reassign the weight value to each particle
Figure BDA00025449502800000416
Obtaining N expected values of the newly generated particles by the formula (8);
Figure BDA00025449502800000417
moreover, the calculating the state parameter by using the extended kalman filter in the step (3) includes the following specific steps:
inputting initial value X of state parameter0＝[a0、α0、β0]And the corresponding variance Q ═ σa,σα,σβ]；
Kalman filtering is the optimal analytic solution of the recursive Bayesian formula, assuming the equation of state Xk+1Observation equation Zk+1The following were used:
Xk+1＝AXk+Wk+1(13)
Zk+1＝HXk+1+Vk+1(14)
in the formula: a is a state transition matrix, Wk+1Predicting noise for a system state; h is an observation matrix, Vk+1To observe noise;
the equations (8) and (10) are generally nonlinear functions, and the core idea of the extended Kalman filter algorithm is to linearize the nonlinear function and to estimate the state of the function Xk+1And an observation function Zk+1Surrounding the filtered value
Figure BDA00025449502800000418
And performing Taylor series expansion and neglecting the influence of second-order and higher-order terms, thereby forming an approximate linearized state parameter estimation model as follows:
Figure BDA0002544950280000051
wherein, it is made
Figure BDA0002544950280000052
Figure BDA0002544950280000053
Figure BDA0002544950280000054
Figure BDA0002544950280000055
The equation (22) and the equation (23) are Jacobian matrixes of the state parameter estimation model and the observation model respectively, and the prediction result of the step k and the equation (21) can be obtained respectively
Figure BDA0002544950280000056
And
Figure BDA0002544950280000057
the extended kalman filter equation can then be expressed as:
Figure BDA0002544950280000058
Figure BDA0002544950280000059
since the predicted value is calculated for the equation (13), the influence of the second order term and higher order terms is ignored after the taylor series expansion, and therefore, the equation (20) has a certain error, and the calculation is performed
Figure BDA00025449502800000510
Calculating by using the formula (13) and the formula (14);
Figure BDA00025449502800000511
in the formula (I), the compound is shown in the specification,
Figure BDA00025449502800000512
predicting value in the k step, and setting the k as 0 to be initial value of state parameter;
after new state parameters are obtained through prediction, the correlation among the state parameters is changed, the system covariance is changed, and the covariance matrix of the prior state estimation is as follows:
Figure BDA00025449502800000513
wherein Q is a system state parameter variance matrix;
and (3) updating: obtaining an observation parameter Zk+1Then, the state prediction value can be corrected, and the correction equation is as the following formula,
Figure BDA0002544950280000061
kalman gain matrix:
Kk+1＝p'k+1HT(Hp'k+1HT+R)-1(24)
and (4) updating the predicted value:
Figure BDA0002544950280000062
covariance update estimation:
pk+1＝(I-Kk+1H)p'k+1(26)。
preferably, the calculating the state parameter by using unscented kalman filtering in step (3) includes the following specific steps:
inputting initial value X of state parameter0＝[a0、α0、β0]And the corresponding variance Q ═ σa,σα,σβ](ii) a Covariance matrix p0；
Obtaining a group of sampling points, a Sigma point set and a corresponding weight value by using a formula (27) and a formula (28); calculating 2n +1 Sigma sampling points and n finger-shaped state function variable dimensions;
Figure BDA0002544950280000063
wherein the content of the first and second substances,
Figure BDA0002544950280000064
column i showing the square root of the matrix;
calculating the weight of the sampling point,
Figure BDA0002544950280000065
wherein, subscript m is mean value, c is covariance;
Figure BDA0002544950280000066
and
Figure BDA0002544950280000067
a weight representing the mean and covariance of the corresponding point, and a parameter λ α2(n + k) -n is a scaling parameter, and the value of the parameter α is [0,1 ]]Determining the distribution of sampling points, and adjusting the sampling points and the mean value
Figure BDA0002544950280000071
K is a candidate parameter for ensuring that the matrix (n + λ) p is a semi-positive definite matrix, k is 0 when n ≧ 3 and n-3 when n < 3, and β is a parameter>0 can adjust equation higher order term dynamic difference, and β is taken as 2 when normal distribution is carried out;
first of all generated to
Figure BDA0002544950280000072
Sigma point set as mean:
Figure BDA0002544950280000073
substituting Sigma point set into state parameter estimation function fk(Xk) Obtaining a set of state parameter estimation sample points
Figure BDA0002544950280000074
Sampling points
Figure BDA0002544950280000075
To correspond to
Figure BDA0002544950280000076
The weighted sum can obtain the mean and variance of the corresponding predicted values of the state parameters, and the specific formula is as follows:
Figure BDA0002544950280000077
Figure BDA0002544950280000078
according to the predicted value
Figure BDA0002544950280000079
Obtaining a set of samples using equations (27) and (28) againSampling points are as follows:
Figure BDA00025449502800000710
the predicted Sigma point set is substituted into observation equation (10) to obtain a predicted observation value, i ═ 1,2, …,2n + 1.
Figure BDA00025449502800000711
Obtaining the mean value and the variance of system prediction through weighted summation according to the obtained prediction observation value;
Figure BDA00025449502800000712
Figure BDA00025449502800000713
Figure BDA00025449502800000714
kalman gain matrix calculation:
Figure BDA00025449502800000715
updating state parameters and covariance to obtain real observation parameters Zk+1Updating the post-state parameters:
Figure BDA0002544950280000081
and (3) covariance updating:
Figure BDA0002544950280000082
further, the step (4) specifically includes the steps of: after the state parameter estimation is carried out by utilizing the step (3), the parameter of the K +1 step can be obtained
Figure BDA0002544950280000083
Predicting the fatigue crack extension life, and obtaining the crack length of the (k + n) th step by only substituting the parameters into a discrete recursion equation to carry out recursion calculation, wherein the smaller the n is, the higher the precision is, certainly, because the state parameters are changed at any time; the concrete prediction formula of the discrete recursive equation is as follows;
Figure BDA0002544950280000084
has the advantages that: compared with the prior art, the invention has the following remarkable advantages: the method aims at the influence of the impact load, environment, structural form, material and other uncertainties on the fatigue crack propagation process in the hoisting process of the ladle crane; establishing an improved double-parameter index model, forming a discretized state parameter evaluation model, and estimating fatigue performance parameters and a crack propagation trend by using a particle filter algorithm, an extended Kalman filter algorithm or an unscented Kalman filter algorithm; finally, predicting the residual life of the fatigue crack of the metal structure of the casting crane according to the crack propagation trend; the comparison shows that the prediction results of the particle filter algorithm, the extended Kalman filter algorithm and the unscented Kalman filter algorithm are superior to those of the prior art, and the maximum values of the relative absolute errors of the predicted values of the extended Kalman filter algorithm, the unscented Kalman filter algorithm and the particle filter algorithm are respectively 2.5%, 2% and 11% when the crack is expanded in the first stage; the prediction requirement of the crack propagation life of the casting crane is met; during the second stage of crack propagation, the maximum values of relative absolute errors of predicted values of an extended Kalman filter algorithm, an unscented Kalman filter algorithm and a particle filter algorithm are respectively 8.5%, 15% and 13.5%; the prediction requirement of the crack propagation life of the casting crane is also met; therefore, the combination of the particle filter algorithm, the extended Kalman filter algorithm and the unscented Kalman filter algorithm with the improved two-parameter index model can well predict the fatigue crack extension life of the metal structure of the casting crane.
Drawings
FIG. 1 is a schematic diagram of a single-beam solid-shell coupling model according to the present invention;
FIG. 2 is a schematic view of a fatigue crack propagation curve of the diaphragm and the lower cover plate according to the present invention;
FIG. 3 is a first stage simulation crack propagation life prediction diagram in accordance with the present invention;
FIG. 4 is a diagram of the second stage simulated crack propagation life prediction in the present invention;
FIG. 5 is a graph of relative absolute error of the first stage of the present invention;
FIG. 6 is a diagram of relative absolute error at the second stage of the present invention;
FIG. 7 is a diagram of the first stage and the true value variance in the present invention;
FIG. 8 is an enlarged schematic view of FIG. 7;
FIG. 9 is a graph of the first stage and mean variance of the present invention;
FIG. 10 is an enlarged schematic view of FIG. 9;
FIG. 11 is a second stage and true variance plot of the present invention;
FIG. 12 is an enlarged schematic view of FIG. 11;
FIG. 13 is a graph of the second stage and mean variance of the present invention;
fig. 14 is an enlarged schematic view of fig. 13.
Detailed Description
The technical scheme of the invention is further explained by combining the attached drawings.
Example 1
The invention discloses a method for predicting the fatigue crack propagation life of a metal structure based on a filtering algorithm, which is characterized by comprising the following steps of:
(1) improving the fatigue crack propagation two-parameter index model to obtain an improved fatigue crack propagation two-parameter index model; wherein the fatigue crack propagation two-parameter exponential model describes the fatigue crack propagation rate, and the formula is as follows:
Figure BDA0002544950280000091
wherein α is the material parameter, γ is the ratio of the loading mode, the crack shape andposition dependent coefficients, Δ σ stress amplitude, a crack length, are obtained from equation (1) when α is a fixed parameter, and β<At 0, the crack propagation rate is increased continuously along with the growth of the crack length, and the increase is less than eαWhen β>0, the crack growth rate decreased with the crack length and was not suitable for life prediction, and it is understood that β<When the crack propagation rate is high, α parameters need to be changed greatly to achieve the prediction effect, and the prediction effect is not beneficial to practical application, so that a two-parameter exponential model is improved as follows:
Figure BDA0002544950280000101
(2) constructing a state space evaluation model based on the improved fatigue crack propagation two-parameter index model;
the method for constructing the state space evaluation model specifically comprises the following steps:
(2.1) constructing a state space parameter evaluation equation:
setting dN as 1, namely the crack propagation length under single cycle load; converting the improved fatigue crack propagation two-parameter exponential model into a discrete recursion form as follows:
Figure BDA0002544950280000102
because the frequent crack parts are dispersed and the stress amplitude on the crack surface is difficult to measure, the delta sigma is replaced by the stress amplitude obtained by finite element simulation; and because the stress amplitude obtained by simulation has calculation error, the stress amplitude delta sigma obtained by simulation is designed to obey normal distribution
Figure BDA0002544950280000103
To simulate the variance of equivalent response amplitude values with true values,
Figure BDA0002544950280000104
is the equivalent stress amplitude;thus ak+1＝f(akΔ σ) in
Figure BDA0002544950280000105
To ak+1Performing first-order Taylor expansion to obtain:
Figure BDA0002544950280000106
let ak+1The process noise of (a) is:
Figure BDA0002544950280000107
in the improved fatigue crack propagation two-parameter exponential model, the partial derivative of the delta sigma is calculated as follows:
Figure BDA0002544950280000108
in that
Figure BDA0002544950280000109
In the case where it is known that,
Figure BDA00025449502800001010
the formula (6) is a definite value, and
Figure BDA00025449502800001011
thus, the
Figure BDA00025449502800001012
σaThe expression of (a) is:
Figure BDA00025449502800001013
the crack length a can be obtained from the formula (1), and when the structural characteristics, environment and working conditions cause the fatigue parameters alpha and beta to change, the crack length a is different; then the state space parameter evaluation equation of the two-parameter exponential model is:
Figure BDA0002544950280000111
in the formula (I), the compound is shown in the specification,
Figure BDA0002544950280000112
white Gaussian noise of α, β, as measured by standard practice fatigue testsα,k、σβ,k、σa,kForming a system state parameter variance matrix Q;
(2.2) constructing a state space observation equation
The state space observation equation obtained from the fatigue crack length obtained by monitoring or detection, the corresponding cycle number and the error existing in the measurement process is as follows:
zk+1＝Hk+1xk+1+vk+1(9)
in the formula, Hk+1Is a measurement matrix, here an identity matrix; x is the number ofk+1Is a measured value, here the crack length ak+1；vk+1To measure errors, and is subject to
Figure BDA0002544950280000113
The state space observation equation is as follows:
Zk+1＝ak+1+vk+1(10)
wherein σr,kForming an observation variance matrix R;
(3) calculating a state parameter X of the state space evaluation model constructed in the step (2) by using a filtering algorithmk+1；
The method for calculating the state parameters by using the particle filter comprises the following specific steps:
initializing state parameters, inputting initial value X of state parameters0＝[a0、α0、β0]And the corresponding variance Q ═ σa,σα,σβ]；
Sampling a set of particles, collecting the set of particles using a probability distribution according to α, a
Figure BDA0002544950280000114
Generating a set of state function samples using a state space parameter estimation equation
Figure BDA0002544950280000115
k +1 is the number of state function steps,
Figure BDA0002544950280000116
is a normal function matrix;
calculating importance weight according to Z in state observation equationk+1Calculating a weight value:
Figure BDA0002544950280000117
wherein the content of the first and second substances,
Figure BDA0002544950280000118
is composed of
Figure BDA0002544950280000119
In the presence of Zk+1The probability of occurrence;
Figure BDA00025449502800001110
is composed of
Figure BDA00025449502800001111
The probability of occurrence;
Figure BDA00025449502800001112
is Zk+1Under the conditions of occurrence
Figure BDA00025449502800001113
The probability of occurrence;
normalizing the weight:
Figure BDA0002544950280000121
random resampling: weighted value according to importance
Figure BDA0002544950280000122
The sample is re-drawn among the N particles,
Figure BDA0002544950280000123
the large number of the distributed particles is more,
Figure BDA0002544950280000124
smaller direct deletion, last by difference
Figure BDA0002544950280000125
Generating new N particles by specific gravity; and reassign the weight value to each particle
Figure BDA0002544950280000126
Obtaining N expected values of the newly generated particles by the formula (8);
Figure BDA0002544950280000127
(4) using the state parameters
Figure BDA0002544950280000128
Predicting the residual life of crack propagation, wherein the method specifically comprises the following steps: after the state parameter estimation is carried out by utilizing the step (3), the parameter of the K +1 step can be obtained
Figure BDA0002544950280000129
Predicting the fatigue crack extension life, and obtaining the crack length of the (k + n) th step by only substituting the parameters into a discrete recursion equation to carry out recursion calculation, wherein the smaller the n is, the higher the precision is, certainly, because the state parameters are changed at any time; the concrete prediction formula of the discrete recursive equation is as follows;
Figure BDA00025449502800001210
example 2
The invention discloses a method for predicting the fatigue crack propagation life of a metal structure based on a filtering algorithm, which is characterized by comprising the following steps of:
(1) improving the fatigue crack propagation two-parameter index model to obtain an improved fatigue crack propagation two-parameter index model; wherein the fatigue crack propagation two-parameter exponential model describes the fatigue crack propagation rate, and the formula is as follows:
Figure BDA00025449502800001211
wherein α is a material parameter, γ is a coefficient related to loading mode, crack shape and position, Δ σ is a stress amplitude, and a is a crack length, and is obtained from formula (1) when α is a fixed parameter, and β is<At 0, the crack propagation rate is increased continuously along with the growth of the crack length, and the increase is less than eαWhen β>0, the crack growth rate decreased with the crack length and was not suitable for life prediction, and it is understood that β<When the crack propagation rate is high, α parameters need to be changed greatly to achieve the prediction effect, and the prediction effect is not beneficial to practical application, so that a two-parameter exponential model is improved as follows:
Figure BDA0002544950280000131
(2) constructing a state space evaluation model based on the improved fatigue crack propagation two-parameter index model;
the method for constructing the state space evaluation model specifically comprises the following steps:
(2.1) constructing a state space parameter evaluation equation:
setting dN as 1, namely the crack propagation length under single cycle load; converting the improved fatigue crack propagation two-parameter exponential model into a discrete recursion form as follows:
Figure BDA0002544950280000132
the frequent parts of the cracks are relatively dispersedAnd the stress amplitude on the surface of the crack is difficult to measure, so that the delta sigma is replaced by the stress amplitude obtained by finite element simulation; and because the stress amplitude obtained by simulation has calculation error, the stress amplitude delta sigma obtained by simulation is designed to obey normal distribution
Figure BDA0002544950280000133
To simulate the variance of equivalent response amplitude values with true values,
Figure BDA0002544950280000134
is the equivalent stress amplitude; thus ak+1＝f(akΔ σ) in
Figure BDA0002544950280000135
Is aligned with ak+1Performing first-order Taylor expansion to obtain:
Figure BDA0002544950280000136
let ak+1The process noise of (a) is:
Figure BDA0002544950280000137
in the improved fatigue crack propagation two-parameter exponential model, the partial derivative of the delta sigma is calculated as follows:
Figure BDA0002544950280000138
in that
Figure BDA0002544950280000139
In the case where it is known that,
Figure BDA00025449502800001310
the formula (6) is a definite value, and
Figure BDA00025449502800001311
thus, the
Figure BDA00025449502800001312
σaThe expression of (a) is:
Figure BDA00025449502800001313
the crack length a can be obtained from the formula (1), and when the structural characteristics, environment and working conditions cause the fatigue parameters alpha and beta to change, the crack length a is different; then the state space parameter evaluation equation of the two-parameter exponential model is:
Figure BDA0002544950280000141
in the formula (I), the compound is shown in the specification,
Figure BDA0002544950280000142
white Gaussian noise of α, β, as measured by standard practice fatigue testsα,k、σβ,k、σa,kForming a system state parameter variance matrix Q;
(2.2) constructing a state space observation equation
The state space observation equation obtained from the fatigue crack length obtained by monitoring or detection, the corresponding cycle number and the error existing in the measurement process is as follows:
zk+1＝Hk+1xk+1+vk+1(9)
in the formula, Hk+1Is a measurement matrix, here an identity matrix; x is the number ofk+1Is a measured value, here the crack length ak+1；vk+1To measure errors, and is subject to
Figure BDA0002544950280000143
The state space observation equation is as follows:
Zk+1＝ak+1+vk+1(10)
wherein σr,kForming an observation variance matrix R;
(3) utilizing a filtering algorithm for the state space evaluation model constructed in the step (2)Calculating a state parameter Xk+1；
The method for calculating the state parameters by using the extended Kalman filter comprises the following specific steps:
inputting initial value X of state parameter0＝[a0、α0、β0]And the corresponding variance Q ═ σa,σα,σβ]；
Kalman filtering is the optimal analytic solution of the recursive Bayesian formula, assuming the equation of state Xk+1Observation equation Zk+1The following were used:
Xk+1＝AXk+Wk+1(13)
Zk+1＝HXk+1+Vk+1(14)
in the formula: a is a state transition matrix, Wk+1Predicting noise for a system state; h is an observation matrix, Vk+1To observe noise;
the equations (8) and (10) are generally nonlinear functions, and the core idea of the extended Kalman filter algorithm is to linearize the nonlinear function and to estimate the state of the function Xk+1And an observation function Zk+1Surrounding the filtered value
Figure BDA0002544950280000144
And performing Taylor series expansion and neglecting the influence of second-order and higher-order terms, thereby forming an approximate linearized state parameter estimation model as follows:
Figure BDA0002544950280000151
wherein, it is made
Figure BDA0002544950280000152
Figure BDA0002544950280000153
Figure BDA0002544950280000154
Figure BDA0002544950280000155
The equation (22) and the equation (23) are Jacobian matrixes of the state parameter estimation model and the observation model respectively, and the prediction result of the step k and the equation (21) can be obtained respectively
Figure BDA0002544950280000156
And
Figure BDA00025449502800001512
the extended kalman filter equation can then be expressed as:
Figure BDA0002544950280000157
Figure BDA0002544950280000158
since the predicted value is calculated for the equation (13), the influence of the second order term and higher order terms is ignored after the taylor series expansion, and therefore, the equation (20) has a certain error, and the calculation is performed
Figure BDA0002544950280000159
Calculating by using the formula (13) and the formula (14);
Figure BDA00025449502800001510
in the formula (I), the compound is shown in the specification,
Figure BDA00025449502800001511
predicting value in the k step, and setting the k as 0 to be initial value of state parameter;
after new state parameters are obtained through prediction, the correlation among the state parameters is changed, the system covariance is changed, and the covariance matrix of the prior state estimation is as follows:
p'k+1＝FKpkFk T+Q (22)
wherein Q is a system state parameter variance matrix;
and (3) updating: obtaining an observation parameter Zk+1Then, the state prediction value can be corrected, and the correction equation is as the following formula,
Figure BDA0002544950280000161
kalman gain matrix:
Kk+1＝p'k+1HT(Hp'k+1HT+R)-1(24)
and (4) updating the predicted value:
Figure BDA0002544950280000162
covariance update estimation:
pk+1＝(I-Kk+1H)p'k+1(26)
(4) using the state parameters
Figure BDA0002544950280000163
Predicting the residual life of crack propagation, wherein the method specifically comprises the following steps: after the state parameter estimation is carried out by utilizing the step (3), the parameter of the K +1 step can be obtained
Figure BDA0002544950280000164
Predicting the fatigue crack extension life, and obtaining the crack length of the (k + n) th step by only substituting the parameters into a discrete recursion equation to carry out recursion calculation, wherein the smaller the n is, the higher the precision is, certainly, because the state parameters are changed at any time; the concrete prediction formula of the discrete recursive equation is as follows;
Figure BDA0002544950280000165
example 3
The invention discloses a method for predicting the fatigue crack propagation life of a metal structure based on a filtering algorithm, which is characterized by comprising the following steps of:
(1) improving the fatigue crack propagation two-parameter index model to obtain an improved fatigue crack propagation two-parameter index model; wherein the fatigue crack propagation two-parameter exponential model describes the fatigue crack propagation rate, and the formula is as follows:
Figure BDA0002544950280000166
wherein α is a material parameter, γ is a coefficient related to loading mode, crack shape and position, Δ σ is a stress amplitude, and a is a crack length, and is obtained from formula (1) when α is a fixed parameter, and β is<At 0, the crack propagation rate is increased continuously along with the growth of the crack length, and the increase is less than eαWhen β>0, the crack growth rate decreased with the crack length and was not suitable for life prediction, and it is understood that β<When the crack propagation rate is high, α parameters need to be changed greatly to achieve the prediction effect, and the prediction effect is not beneficial to practical application, so that a two-parameter exponential model is improved as follows:
Figure BDA0002544950280000171
(2) constructing a state space evaluation model based on the improved fatigue crack propagation two-parameter index model;
the method for constructing the state space evaluation model specifically comprises the following steps:
(2.1) constructing a state space parameter evaluation equation:
setting dN as 1, namely the crack propagation length under single cycle load; converting the improved fatigue crack propagation two-parameter exponential model into a discrete recursion form as follows:
Figure BDA0002544950280000172
the frequent parts of the cracks are more dispersed, and the surfaces of the cracks are stressedThe measurement of the force amplitude is difficult, so that the delta sigma is replaced by the stress amplitude obtained by finite element simulation; and because the stress amplitude obtained by simulation has calculation error, the stress amplitude delta sigma obtained by simulation is designed to obey normal distribution
Figure BDA0002544950280000173
To simulate the variance of equivalent response amplitude values with true values,
Figure BDA0002544950280000174
is an equivalent stress magnitude, thus αk+1＝f(akΔ σ) in
Figure BDA0002544950280000175
Is aligned with ak+1Performing first-order Taylor expansion to obtain:
Figure BDA0002544950280000176
let ak+1The process noise of (a) is:
Figure BDA0002544950280000177
in the improved fatigue crack propagation two-parameter exponential model, the partial derivative of the delta sigma is calculated as follows:
Figure BDA0002544950280000178
in that
Figure BDA0002544950280000179
In the case where it is known that,
Figure BDA00025449502800001710
and formula (6) are both definite values, and
Figure BDA00025449502800001711
thus, the
Figure BDA0002544950280000181
σaThe expression of (a) is:
Figure BDA0002544950280000182
the crack length a can be obtained from the formula (1), and when the structural characteristics, environment and working conditions cause the fatigue parameters alpha and beta to change, the crack length a is different; then the state space parameter evaluation equation of the two-parameter exponential model is:
Figure BDA0002544950280000183
in the formula (I), the compound is shown in the specification,
Figure BDA0002544950280000184
white Gaussian noise of α, β, as measured by standard practice fatigue testsα,k、σβ,k、σa,kForming a system state parameter variance matrix Q;
(2.2) constructing a state space observation equation
The state space observation equation obtained from the fatigue crack length obtained by monitoring or detection, the corresponding cycle number and the error existing in the measurement process is as follows:
zk+1＝Hk+1xk+1+vk+1(9)
in the formula, Hk+1Is a measurement matrix, here an identity matrix; x is the number ofk+1Is a measured value, here the crack length ak+1；vk+1To measure errors, and is subject to
Figure BDA0002544950280000185
The state space observation equation is as follows:
Zk+1＝ak+1+vk+1(10)
wherein σr,kForming an observation variance matrix R;
(3) calculating a state parameter X of the state space evaluation model constructed in the step (2) by using a filtering algorithmk+1；
The method for calculating the state parameters by using the unscented Kalman filtering comprises the following specific steps:
inputting initial value X of state parameter0＝[a0、α0、β0]And the corresponding variance Q ═ σa,σα,σβ](ii) a Covariance matrix p0；
Obtaining a group of sampling points, a Sigma point set and a corresponding weight value by using a formula (27) and a formula (28); calculating 2n +1 Sigma sampling points and n finger-shaped state function variable dimensions;
Figure BDA0002544950280000191
wherein the content of the first and second substances,
Figure BDA0002544950280000192
the ith column representing the square root of the matrix;
calculating the weight of the sampling point,
Figure BDA0002544950280000193
wherein, subscript m is mean value, c is covariance;
Figure BDA0002544950280000194
and
Figure BDA0002544950280000195
a weight representing the mean and covariance of the corresponding point, and a parameter λ α2(n + k) -n is a scaling parameter, and the value of the parameter α is [0,1 ]]Determining the distribution of sampling points, and adjusting the sampling points and the mean value
Figure BDA0002544950280000196
K is a candidate parameter for ensuring that the matrix (n + λ) p is a semi-positive definite matrix, k is 0 when n ≧ 3 and n-3 when n < 3, and β is a parameter>0 can adjust equation higher order term dynamic difference, and β is taken as 2 when normal distribution is carried out;
first of all generated to
Figure BDA0002544950280000197
Sigma point set as mean:
Figure BDA0002544950280000198
substituting Sigma point set into state parameter estimation function fk(Xk) Obtaining a set of state parameter estimation sample points
Figure BDA0002544950280000199
Sampling points
Figure BDA00025449502800001910
Corresponding to
Figure BDA00025449502800001911
The weighted sum can obtain the mean and variance of the corresponding predicted values of the state parameters, and the specific formula is as follows:
Figure BDA00025449502800001912
Figure BDA00025449502800001913
according to the predicted value
Figure BDA00025449502800001914
A set of samples is obtained again using equations (27) and (28):
Figure BDA00025449502800001915
the predicted Sigma point set is substituted into observation equation (10) to obtain a predicted observation value, i ═ 1,2, …,2n + 1.
Figure BDA0002544950280000201
Obtaining the mean value and the variance of system prediction through weighted summation according to the obtained prediction observation value;
Figure BDA0002544950280000202
Figure BDA0002544950280000203
Figure BDA0002544950280000204
kalman gain matrix calculation:
Figure BDA0002544950280000205
updating state parameters and covariance to obtain real observation parameters Zk+1Updating the post-state parameters:
Figure BDA0002544950280000206
and (3) covariance updating:
Figure BDA0002544950280000207
(4) using the state parameters
Figure BDA0002544950280000208
Predicting the residual life of crack propagation, wherein the method specifically comprises the following steps: after the state parameter estimation is carried out by utilizing the step (3), the parameter of the K +1 step can be obtained
Figure BDA0002544950280000209
Predicting the fatigue crack extension life, and obtaining the crack length of the (k + n) th step by only substituting the parameters into a discrete recursion equation to carry out recursion calculation, wherein the smaller the n is, the higher the precision is, certainly, because the state parameters are changed at any time; discrete recursionThe concrete prediction formula of the equation is as follows;
Figure BDA00025449502800002010
in order to verify the accuracy of the prediction method of the above embodiment, the crack propagation life is predicted according to the result of the Abaqus simulation shown in fig. 1 and 2.
Initial crack length a in improved two-parameter index model0The geometry factor γ is 1.125 at 16mm, and the assumptions of the initial distribution are as in table 1 below.
TABLE 1 simulation parameter set based on improved two-parameter exponential model
Figure BDA0002544950280000211
Note: delta sigma is the equivalent stress amplitude; alpha is a fatigue parameter; beta is a fatigue parameter; and N is a normal distribution function.
The simulated crack life prediction is as shown in fig. 3 and 4, and the parameter estimation of the crack propagation in the first stage and the second stage and the crack propagation life prediction are accurate.
Selecting evaluation parameters obtained in the first and second stages of ABAQUS simulation results under the cycle times (41000, 42000, 43000, 44000, 45000, 46000, 47000, 48000, 49000, 50000, 61000, 62000, 63000 and 64000)
Figure BDA0002544950280000212
AREk+1、Vk+1、
Figure BDA0002544950280000213
And comparing and analyzing the advantages and disadvantages of the three methods in crack life prediction. The respective estimated values, the mean, the relative absolute error, and the variance of the predicted values are shown in fig. 5 to 14.
As shown in fig. 5 and 6, in the parameter evaluation stage, the crack propagation length can be well described by combining the improved biparametric index formula with the extended kalman filter algorithm, the unscented kalman filter algorithm and the particle filter algorithm, and during the crack propagation in the first stage, the maximum values of the relative absolute errors of the predicted values of the extended kalman filter algorithm, the unscented kalman filter algorithm and the particle filter algorithm are respectively 2.5%, 2% and 11%; the prediction requirement of the crack propagation life of the casting crane is met; during the second stage of crack propagation, the maximum values of relative absolute errors of predicted values of an extended Kalman filter algorithm, an unscented Kalman filter algorithm and a particle filter algorithm are respectively 8.5%, 15% and 13.5%; and the prediction requirement of the crack propagation life of the casting crane is also met. And when the service life prediction order is reached, the crack prediction by the extended Kalman filtering algorithm and the unscented Kalman filtering algorithm is more accurate than that by the particle filtering algorithm.
Three algorithms were evaluated:
1) as shown in fig. 7 and 8, when the crack propagates in the first stage, the estimation results of the unscented kalman filter algorithm and the particle filter algorithm are both better than those of the extended kalman filter algorithm. In the prediction stage, the variance of the extended Kalman filtering algorithm is smaller than that of the particle filtering algorithm and the unscented Kalman filtering algorithm. Therefore, in terms of result accuracy, the first-stage life prediction method uses an extended Kalman filtering and improved two-parameter exponential model combined algorithm.
2) As shown in fig. 9 and 10, in the first stage of crack propagation prediction, the convergence of the particle filter algorithm and the extended kalman filter algorithm is better than that of the unscented kalman filter algorithm. In the parameter estimation stage, the convergence of unscented Kalman filtering and particle algorithm is superior to that of the extended Kalman filtering algorithm, but the service life prediction of the extended Kalman filtering algorithm is not influenced by the small divergence of the extended Kalman filtering algorithm.
3) As shown in fig. 11 and 12, in the second stage of crack propagation, in the parameter estimation stage, the estimated value of the unscented kalman filter algorithm is closest to the true value. And in the life prediction stage, the predicted value of the particle filter algorithm is closest to the real result, then the extended Kalman filter algorithm is used, and finally the unscented Kalman filter algorithm is used.
4) As shown in fig. 13 and 14, in the second stage of crack propagation, in the parameter estimation stage, the unscented kalman filter algorithm has the best convergence, and in the life prediction stage, the predicted value of the particle filter algorithm has the best convergence, and then the extended kalman filter algorithm and finally the unscented kalman filter algorithm are performed.
In summary, when the improved two-parameter index model is fused with the filter algorithm, when the crack propagation rate is slow and the crack length is short (i.e. the crack propagation in the first stage), the prediction result of the extended kalman filter algorithm is superior to that of the particle filter algorithm and the unscented kalman filter algorithm. The crack propagation speed is high, the crack length is long (namely, the crack propagation in the second stage), and the prediction result is superior to the unscented Kalman filter algorithm and the extended Kalman filter algorithm by adopting the particle filter algorithm.

Claims (7)

1. A metal structure fatigue crack propagation life prediction method based on a filtering algorithm is characterized by comprising the following steps:
(1) improving the fatigue crack propagation two-parameter index model to obtain an improved fatigue crack propagation two-parameter index model;
(2) constructing a state space evaluation model based on the improved fatigue crack propagation two-parameter index model;
(3) calculating a state parameter X of the state space evaluation model constructed in the step (2) by using a filtering algorithmk+1；
(4) Using the state parameters
Figure FDA0002544950270000011
And predicting the residual life of the crack propagation.
2. The method for predicting the fatigue crack propagation life of the metal structure based on the filter algorithm is characterized by comprising the following steps of: the two-parameter exponential model of fatigue crack propagation in the step (1) describes the fatigue crack propagation rate, and the formula is as follows:
Figure FDA0002544950270000012
wherein α is a material parameter, and γ is a system related to the loading manner, crack shape and positionNumber, Δ σ is the stress amplitude, a is the crack length, as can be obtained from equation (1) when α is a fixed parameter, and β<At 0, the crack propagation rate is increased continuously along with the growth of the crack length, and the increase is less than eαWhen β>0, the crack growth rate decreased with the crack length and was not suitable for life prediction, and it is understood that β<When the crack propagation rate is high, α parameters need to be changed greatly to achieve the prediction effect, and the prediction effect is not beneficial to practical application, so that a two-parameter exponential model is improved as follows:
Figure FDA0002544950270000013
3. the method for predicting the fatigue crack propagation life of the metal structure based on the filter algorithm is characterized by comprising the following steps of: the method for constructing the state space evaluation model in the step (2) specifically comprises the following steps:
(2.1) constructing a state space parameter evaluation equation:
setting dN as 1, namely the crack propagation length under single cycle load; converting the improved fatigue crack propagation two-parameter exponential model into a discrete recursion form as follows:
Figure FDA0002544950270000014
because the frequent crack parts are dispersed and the stress amplitude on the crack surface is difficult to measure, the delta sigma is replaced by the stress amplitude obtained by finite element simulation; and because the stress amplitude obtained by simulation has calculation error, the stress amplitude delta sigma obtained by simulation is designed to obey normal distribution
Figure FDA0002544950270000021
Figure FDA0002544950270000022
To simulate the variance of equivalent response amplitude values with true values,
Figure FDA0002544950270000023
is an equivalent stress magnitude, thus αk+1＝f(akΔ σ) in
Figure FDA0002544950270000024
Is aligned with ak+1Performing first-order Taylor expansion to obtain:
Figure FDA0002544950270000025
let ak+1The process noise of (a) is:
Figure FDA0002544950270000026
in the improved fatigue crack propagation two-parameter exponential model, the partial derivative of the delta sigma is calculated as follows:
Figure FDA0002544950270000027
in that
Figure FDA0002544950270000028
In the case where it is known that,
Figure FDA0002544950270000029
and formula (6) are both definite values, and
Figure FDA00025449502700000210
thus, the
Figure FDA00025449502700000211
σaThe expression of (a) is:
Figure FDA00025449502700000212
the crack length a can be obtained from the formula (1), and when the structural characteristics, environment and working conditions cause the fatigue parameters alpha and beta to change, the crack length a is different; then the state space parameter evaluation equation of the two-parameter exponential model is:
Figure FDA00025449502700000213
in the formula (I), the compound is shown in the specification,
Figure FDA00025449502700000214
white Gaussian noise of α, β, as measured by standard practice fatigue testsα,k、σβ,k、σa,kForming a system state parameter variance matrix Q;
(2.2) constructing a state space observation equation
The state space observation equation obtained from the fatigue crack length obtained by monitoring or detection, the corresponding cycle number and the error existing in the measurement process is as follows:
zk+1＝Hk+1xk+1+vk+1(9)
in the formula, Hk+1Is a measurement matrix, here an identity matrix; x is the number ofk+1Is a measured value, here the crack length ak+1；vk+1To measure errors, and is subject to
Figure FDA0002544950270000031
The state space observation equation is as follows:
Zk+1＝ak+1+vk+1(10)
wherein σr,kForming an observation variance matrix R.
4. The method for predicting the fatigue crack propagation life of the metal structure based on the filter algorithm is characterized by comprising the following steps of: the step (3) of calculating the state parameters by using particle filtering comprises the following specific steps:
initializing state parameters, inputting initial value X of state parameters0＝[a0、α0、β0]And the corresponding variance Q ═ σa,σα,σβ]；
Sampling a set of particles, collecting the set of particles using a probability distribution according to α, a
Figure FDA0002544950270000032
Generating a set of state function samples using a state space parameter estimation equation
Figure FDA0002544950270000033
k +1 is the number of state function steps,
Figure FDA0002544950270000034
is a normal function matrix;
calculating importance weight according to Z in state observation equationk+1Calculating a weight value:
Figure FDA0002544950270000035
wherein the content of the first and second substances,
Figure FDA0002544950270000036
is composed of
Figure FDA0002544950270000037
In the presence of Zk+1The probability of occurrence;
Figure FDA0002544950270000038
is composed of
Figure FDA0002544950270000039
The probability of occurrence;
Figure FDA00025449502700000310
is Zk+1Under the conditions of occurrence
Figure FDA00025449502700000311
The probability of occurrence;
normalizing the weight:
Figure FDA00025449502700000312
random resampling: weighted value according to importance
Figure FDA00025449502700000313
The sample is re-drawn among the N particles,
Figure FDA00025449502700000314
the large number of the distributed particles is more,
Figure FDA00025449502700000315
smaller direct deletion, last by difference
Figure FDA00025449502700000316
Generating new N particles by specific gravity; and reassign the weight value to each particle
Figure FDA00025449502700000317
Obtaining N expected values of the newly generated particles by the formula (8);
Figure FDA00025449502700000318
5. the method for predicting the fatigue crack propagation life of the metal structure based on the filter algorithm is characterized by comprising the following steps of: the step (3) of calculating the state parameters by using the extended kalman filter includes the following specific steps:
inputting initial value X of state parameter0＝[a0、α0、β0]And the corresponding variance Q ═ 2σa,σα,σβ]；
Kalman filtering is the optimal analytic solution of the recursive Bayesian formula, assuming the equation of state Xk+1Observation equation Zk+1The following were used:
Xk+1＝AXk+Wk+1(13)
Zk+1＝HXk+1+Vk+1(14)
in the formula: a is a state transition matrix, Wk+1Predicting noise for a system state; h is an observation matrix, Vk+1To observe noise;
the equations (8) and (10) are generally nonlinear functions, and the core idea of the extended Kalman filter algorithm is to linearize the nonlinear function and to estimate the state of the function Xk+1And an observation function Zk+1Surrounding the filtered value
Figure FDA0002544950270000041
And performing Taylor series expansion and neglecting the influence of second-order and higher-order terms, thereby forming an approximate linearized state parameter estimation model as follows:
Figure FDA0002544950270000042
wherein, it is made
Figure FDA0002544950270000043
Figure FDA0002544950270000044
Figure FDA0002544950270000045
Figure FDA0002544950270000046
The equation (16) and the equation (17) are Jacobian matrixes of the state parameter estimation model and the observation model respectively, and the prediction result of the step k and the equation (21) can be obtained respectively
Figure FDA0002544950270000047
And
Figure FDA0002544950270000048
the extended kalman filter equation can then be expressed as:
Figure FDA0002544950270000051
Figure FDA0002544950270000052
since the predicted value is calculated for the equation (13), the influence of the second order term and higher order terms is ignored after the taylor series expansion, and therefore, the equation (20) has a certain error, and the calculation is performed
Figure FDA0002544950270000053
Calculating by using the formula (13) and the formula (14);
Figure FDA0002544950270000054
in the formula (I), the compound is shown in the specification,
Figure FDA0002544950270000055
predicting value in the k step, and setting the k as 0 to be initial value of state parameter;
after new state parameters are obtained through prediction, the correlation among the state parameters is changed, the system covariance is changed, and the covariance matrix of the prior state estimation is as follows:
Figure FDA0002544950270000056
wherein Q is a system state parameter variance matrix;
and (3) updating:obtaining an observation parameter Zk+1The predicted state value can then be corrected, and the correction equation is as follows:
Figure FDA0002544950270000057
kalman gain matrix:
Kk+1＝p'k+1HT(Hp'k+1HT+R)-1(24)
and (4) updating the predicted value:
Figure FDA0002544950270000058
covariance update estimation:
pk+1＝(I-Kk+1H)p'k+1(26)。
6. the method for predicting the fatigue crack propagation life of the metal structure based on the filter algorithm is characterized by comprising the following steps of: the calculating of the state parameters by using the unscented kalman filter in the step (3) comprises the following specific steps:
inputting initial value X of state parameter0＝[a0、α0、β0]And the corresponding variance Q ═ σa,σα,σβ](ii) a Covariance matrix p0；
Obtaining a group of sampling points, a Sigma point set and a corresponding weight value by using a formula (27) and a formula (28); calculate 2n +1 Sigma sampling points, n finger function variable dimensions:
Figure FDA0002544950270000061
wherein the content of the first and second substances,
Figure FDA0002544950270000062
Figure FDA0002544950270000063
the ith column representing the square root of the matrix;
calculating the weight of the sampling point,
Figure FDA0002544950270000064
wherein, subscript m is mean value, c is covariance;
Figure FDA0002544950270000065
and
Figure FDA0002544950270000066
a weight representing the mean and covariance of the corresponding point, and a parameter λ α2(n + k) -n is a scaling parameter, and the value of the parameter α is [0,1 ]]Determining the distribution of sampling points, and adjusting the sampling points and the mean value
Figure FDA0002544950270000067
K is a candidate parameter for ensuring that the matrix (n + λ) p is a semi-positive definite matrix, k is 0 when n ≧ 3 and n-3 when n < 3, and β is a parameter>0 can adjust equation higher order term dynamic difference, and β is taken as 2 when normal distribution is carried out;
first of all generated to
Figure FDA0002544950270000068
Sigma point set as mean:
Figure FDA0002544950270000069
substituting Sigma point set into state parameter estimation function fk(Xk) Obtaining a set of state parameter estimation sample points
Figure FDA00025449502700000610
Sampling points
Figure FDA00025449502700000611
To correspond to
Figure FDA00025449502700000612
The weighted sum can obtain the mean and variance of the corresponding predicted values of the state parameters, and the specific formula is as follows:
Figure FDA00025449502700000613
Figure FDA00025449502700000614
according to the predicted value
Figure FDA00025449502700000615
A set of samples is obtained again using equations (27) and (28):
Figure FDA00025449502700000616
the predicted Sigma point set is substituted into observation equation (10) to obtain a predicted observation value, i ═ 1,2, …,2n + 1.
Figure FDA0002544950270000071
Obtaining the mean value and the variance of system prediction through weighted summation according to the obtained prediction observation value;
Figure FDA0002544950270000072
Figure FDA0002544950270000073
Figure FDA0002544950270000074
kalman gain matrix calculation:
Figure FDA0002544950270000075
updating state parameters and covariance to obtain real observation parameters Zk+1Updating the post-state parameters:
Figure FDA0002544950270000076
and (3) covariance updating:
Figure FDA0002544950270000077
7. the method for predicting the fatigue crack propagation life of the metal structure based on the filter algorithm according to any one of claims 4 to 6, wherein: the step (4) specifically comprises the following steps: after the state parameter estimation is carried out by utilizing the step (3), the parameter of the K +1 step can be obtained
Figure FDA0002544950270000078
Predicting the fatigue crack extension life, and obtaining the crack length of the (k + n) th step by only substituting the parameters into a discrete recursion equation to carry out recursion calculation, wherein the smaller the n is, the higher the precision is, certainly, because the state parameters are changed at any time; the concrete prediction formula of the discrete recursive equation is as follows;
Figure FDA0002544950270000079
CN202010557546.2A 2020-06-18 2020-06-18 Metal structure fatigue crack propagation life prediction method based on filtering algorithm Active CN111783243B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010557546.2A CN111783243B (en) 2020-06-18 2020-06-18 Metal structure fatigue crack propagation life prediction method based on filtering algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010557546.2A CN111783243B (en) 2020-06-18 2020-06-18 Metal structure fatigue crack propagation life prediction method based on filtering algorithm

Publications (2)

Publication Number Publication Date
CN111783243A true CN111783243A (en) 2020-10-16
CN111783243B CN111783243B (en) 2021-02-26

Family

ID=72756957

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010557546.2A Active CN111783243B (en) 2020-06-18 2020-06-18 Metal structure fatigue crack propagation life prediction method based on filtering algorithm

Country Status (1)

Country Link
CN (1) CN111783243B (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112380705A (en) * 2020-11-16 2021-02-19 江南大学 Metal fatigue crack propagation prediction method based on nonlinear prediction filtering algorithm
CN112699579A (en) * 2020-12-09 2021-04-23 电子科技大学 IGBT state monitoring method based on finite element simulation
CN112905953A (en) * 2021-03-22 2021-06-04 东南大学 Unscented Kalman filtering control method for inverted pendulum
CN112989659A (en) * 2021-03-10 2021-06-18 北京航空航天大学 Method for establishing surface crack strength factor database based on point weight function method
CN113065224A (en) * 2021-03-05 2021-07-02 天津大学 Deep sea pipeline crack propagation monitoring and reliability evaluation method based on image recognition
CN113204861A (en) * 2021-03-18 2021-08-03 北京航空航天大学 Metal plate crack damage evaluation system and method
CN113378420A (en) * 2020-11-27 2021-09-10 东南大学 Method for predicting service life of crack pipeline
CN113591340A (en) * 2021-06-24 2021-11-02 内蒙古工业大学 Gear bending fatigue time-varying reliability analysis method under random load
CN113610812A (en) * 2021-08-10 2021-11-05 哈尔滨理工大学 Weight calculation method for judging and influencing factors of subsurface crack form of ultrasonic vibration grinding optical glass
CN113987900A (en) * 2021-10-26 2022-01-28 电子科技大学 IGBT service life prediction method based on extended Kalman particle filter

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102507895A (en) * 2011-11-01 2012-06-20 东南大学 Fatigue life estimation method of cast steel joint with obvious defect in cast steel joints
CN106886663A (en) * 2017-03-29 2017-06-23 北京理工大学 Tooth bending Prediction method for fatigue life and device
CN109165407A (en) * 2018-07-18 2019-01-08 上海工程技术大学 A kind of predictor method for the mechanical component fatigue crack service life
CN109345032A (en) * 2018-10-30 2019-02-15 南京航空航天大学 Particle filter multiple cracks based on Dynamic Crack number extend prediction technique
CN110135085A (en) * 2019-05-20 2019-08-16 北京航空航天大学 Fatigue crack appraisal procedure based on self-adaptive kernel density estimation Auxiliary Particle Filter
CN110489848A (en) * 2019-08-12 2019-11-22 武汉钢铁有限公司 A kind of difference seawater velocity Corrosion Fatigue Crack Propagation rate prediction method

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102507895A (en) * 2011-11-01 2012-06-20 东南大学 Fatigue life estimation method of cast steel joint with obvious defect in cast steel joints
CN106886663A (en) * 2017-03-29 2017-06-23 北京理工大学 Tooth bending Prediction method for fatigue life and device
CN109165407A (en) * 2018-07-18 2019-01-08 上海工程技术大学 A kind of predictor method for the mechanical component fatigue crack service life
CN109345032A (en) * 2018-10-30 2019-02-15 南京航空航天大学 Particle filter multiple cracks based on Dynamic Crack number extend prediction technique
CN110135085A (en) * 2019-05-20 2019-08-16 北京航空航天大学 Fatigue crack appraisal procedure based on self-adaptive kernel density estimation Auxiliary Particle Filter
CN110489848A (en) * 2019-08-12 2019-11-22 武汉钢铁有限公司 A kind of difference seawater velocity Corrosion Fatigue Crack Propagation rate prediction method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
JIAN CHEN: "Research on a Lamb Wave and Particle Filter-Based On-Line Crack Propagation Prognosis Method", 《SENSORS》 *

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112380705B (en) * 2020-11-16 2024-02-06 江南大学 Metal fatigue crack growth prediction method based on nonlinear prediction filtering algorithm
CN112380705A (en) * 2020-11-16 2021-02-19 江南大学 Metal fatigue crack propagation prediction method based on nonlinear prediction filtering algorithm
CN113378420A (en) * 2020-11-27 2021-09-10 东南大学 Method for predicting service life of crack pipeline
CN112699579A (en) * 2020-12-09 2021-04-23 电子科技大学 IGBT state monitoring method based on finite element simulation
CN113065224A (en) * 2021-03-05 2021-07-02 天津大学 Deep sea pipeline crack propagation monitoring and reliability evaluation method based on image recognition
CN113065224B (en) * 2021-03-05 2022-05-17 天津大学 Deep sea pipeline crack propagation monitoring and reliability evaluation method based on image recognition
CN112989659A (en) * 2021-03-10 2021-06-18 北京航空航天大学 Method for establishing surface crack strength factor database based on point weight function method
CN112989659B (en) * 2021-03-10 2022-03-04 北京航空航天大学 Method for establishing surface crack strength factor database based on point weight function method
CN113204861A (en) * 2021-03-18 2021-08-03 北京航空航天大学 Metal plate crack damage evaluation system and method
CN112905953A (en) * 2021-03-22 2021-06-04 东南大学 Unscented Kalman filtering control method for inverted pendulum
CN113591340A (en) * 2021-06-24 2021-11-02 内蒙古工业大学 Gear bending fatigue time-varying reliability analysis method under random load
CN113591340B (en) * 2021-06-24 2023-09-01 内蒙古工业大学 Gear bending fatigue time-varying reliability analysis method under random load
CN113610812A (en) * 2021-08-10 2021-11-05 哈尔滨理工大学 Weight calculation method for judging and influencing factors of subsurface crack form of ultrasonic vibration grinding optical glass
CN113610812B (en) * 2021-08-10 2024-02-02 哈尔滨理工大学 Weight resolving method for judging subsurface crack form of ultrasonic vibration grinding optical glass and influencing factors
CN113987900A (en) * 2021-10-26 2022-01-28 电子科技大学 IGBT service life prediction method based on extended Kalman particle filter

Also Published As

Publication number Publication date
CN111783243B (en) 2021-02-26

Similar Documents

Publication Publication Date Title
CN111783243B (en) Metal structure fatigue crack propagation life prediction method based on filtering algorithm
CN109725123B (en) Crack propagation life determination method considering shot peening strengthening surface layer grain refinement
CN111832125B (en) Method for predicting fatigue crack propagation life of metal structure of metallurgical crane
Lu et al. Small time scale fatigue crack growth analysis
CN104833536A (en) Structure fatigue life calculation method based on non-linear cumulative damage theory
CN111651924B (en) Weight function-based surface crack stress intensity factor database establishment method
Sankararaman et al. Statistical inference of equivalent initial flaw size with complicated structural geometry and multi-axial variable amplitude loading
CN112487683B (en) Structural member high-cycle fatigue life prediction method considering residual stress influence
CN113065702B (en) Landslide displacement multi-linear prediction method based on ST-SEEP segmentation method and space-time ARMA model
CN110987676A (en) Full-life prediction method considering crack closure effect under random multi-axis load
CN112380705A (en) Metal fatigue crack propagation prediction method based on nonlinear prediction filtering algorithm
CN117594164A (en) Metal structure residual fatigue life calculation and evaluation method and system based on digital twin
CN112926698B (en) Vibration prediction and assembly evaluation method for large-scale rotating equipment
US9081922B2 (en) Method of fabricating a mechanical part, including a method of predicting the risks of crack initiation in the part in a “fretting-fatigue” situation
CN113360983B (en) Slope reliability analysis and risk assessment method
CN111044351A (en) DIC technology-based weld joint creep deformation prediction method
WO2021032906A1 (en) A system and a method for monitoring material fatigue
JP2013096862A (en) Crack extension behavior prediction method
CN109766637B (en) Bridge crane structure reliability optimization method based on Krigng agent model
CN108062435B (en) Fatigue life calibration method based on nominal stress method
CN114492074A (en) Probabilistic damage tolerance assessment analysis method
CN116050202A (en) Chaboche blade fatigue life prediction model and method suitable for various average stress performances
CN115482888A (en) Method for predicting crack propagation life under action of pressure-pressure cyclic load
CN108169013A (en) A kind of fatigue life calibration method suitable for Multi-fasteners joint test specimen
CN113283137A (en) Structural member corrosion effect control simulation method under organic protective coating system

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
GR01 Patent grant
GR01 Patent grant