CN104897769A - Magnetic flux leakage testing defect reconstruction method based on improved particle swarm optimization algorithm - Google Patents

Abstract

The invention relates to a magnetic flux leakage testing defect reconstruction method based on an improved particle swarm optimization algorithm. According to the method, self-adaption mutagenic factors are introduced into a particle swarm EPUS-PSO algorithm based on an effective population utilization strategy, thus obtaining the IEPUS-PSO algorithm, and then the IEPUS-PSO algorithm is applied to defect reconstruction in magnetic flux leakage testing. With the improved algorithm, the reconstruction accuracy is improved, the computation time is reduced, and the defect outline can be reconstructed by magnetic flux leakage signals for defects of different sizes.

Description

Based on the Magnetic Flux Leakage Inspecting Root cause analysis method of the particle swarm optimization algorithm improved

Technical field

The present invention relates to a kind of Magnetic Flux Leakage Inspecting technology, in particular to a kind of Magnetic Flux Leakage Inspecting Root cause analysis method of effective population Utilization strategies particle group optimizing (Improved Efficient Population Utilization Strategy for Particle Swarm Optimization, the IEPUS-PSO) algorithm based on improving.

Background technology

In the last few years, the economic development of China, industrial scale constantly expanded, and electrical production has become the mainstay industry of China.Therefore, power equipment production and in safeguarding the application of Dynamic Non-Destruction Measurement also more and more receive publicity.As a kind of detection method the most frequently used in Non-Destructive Testing, Magnetic Flux Leakage Inspecting is more extensive in the application in the fields such as iron and steel, oil, petrochemical industry, its mainly to ferrimagnet surface and nearly surface corrosion, crackle, pore, pit, the detection of defect such as to be mingled with, also can be used for the thickness measuring of ferrimagnet.

Defect profile reconstruct can express the information of defect vividerly, and for the leakage field reconstructing method of a superperformance, the suitable model that moves ahead is necessary, and an optimization problem can be considered as the defect profile reconstruct of a given forward model, because profile is made up of multiple uniform discrete value usually, this generates the dimension of this problem, solve the profile reconstruction that these discrete values are formed, be equivalent to solve a high-dimensional optimization.

PSO is as a kind of powerful stochastic evolutionary algorithm, can be used for finding the globally optimal solution in complex search space, but when solving higher-dimension practical problems due to the complex nature of the problem, easily be absorbed in local optimum too early, can not the actual profile of reconstruction defect exactly, therefore a kind ofly can avoid being absorbed in local optimum and the improvement PSO algorithm that can solve high-dimensional optimization in the urgent need to finding.

Summary of the invention

The present invention be directed to how improvement strategy particle swarm optimization algorithm and be applied to pipeline magnetic flux leakage defect reconstruct problem, propose a kind of Magnetic Flux Leakage Inspecting Root cause analysis method of the particle swarm optimization algorithm based on improving, in the EPUS-PSO algorithm that the TSP question factor is incorporated into, obtain IEPUS-PSO algorithm of the present invention, and by the Root cause analysis of IEPUS-PSO algorithm application in Magnetic Flux Leakage Inspecting, the algorithm after improvement can improve reconstruction accuracy and reduce computing time.

Technical scheme of the present invention is: a kind of Magnetic Flux Leakage Inspecting Root cause analysis method of the particle swarm optimization algorithm based on improving, is characterized in that, specifically comprise the steps:

1): the parameter that effective population Utilization strategies particle group optimizing IEPUS-PSO algorithm is set, comprise maximum iteration time iteration, primary number and maximum number of particles, solution space dimension and solution space scope, particle position represents the profile of defect;

2): set up fitness function wherein d is the dimension of particle, and D is total dimension of particle, p _dthe prediction magnetic leakage signal of forward model, y _dbe actual measurement magnetic leakage signal, iterations s=1 is set;

3): judge that particle adopts hunting zone sharing policy still to separate sharing policy, when activating probability P r (s) and being less than the random number of 0 to 1, then adopt hunting zone sharing policy, otherwise then adopt solution sharing policy, the formula of Pr (s) is:

$\mathrm{Pr}\left(s\right)=0.03+\frac{0.07*s}{\mathrm{iteration}}$

Wherein iteration is maximum iteration time, and s is current iteration number of times;

: the fitness value calculating all particles according to fitness function, and upgrade current individual optimum solution and the globally optimal solution of particle colony 4);

5): population management device adjusts population size according to IEPUS-PSO algorithm, effectively changed the number of particle by the fitness value change of the globally optimal solution of colony, specific rules is:

If a the fitness value of () globally optimal solution does not all upgrade in double iteration, then increase a particle in colony, its place value is:

$x_{\mathrm{add}}=\frac{\mathrm{Pbest}\left(a_1\right)+\mathrm{Pbest}\left(a_2\right)}{2}$

Wherein a ₁and a ₂the sequence number of two particles is randomly drawed in representative from current group, Pbest (a ₁) and Pbest (a ₂) be the current individual optimum solution of extracted two particles, if after adding this particle, when number of particles is greater than set maximum population, needs first to remove a particle that fitness value is the poorest, then add this new particle;

If b the fitness value of () globally optimal solution is all upgraded in double iteration, then that the poorest for fitness value particle enough, is then removed by the number that particle is described;

6): utilize mutagenic factor to carry out disturbance to the position of all particles, the formula of mutagenic factor is:

X' _i,j=x _i,j+ b*rand, x _i,jbe the original position of the jth dimension of i-th particle, i and j is sequence number and the dimension of particle respectively, and rand is the random number between 0 to 1, x' _i,jfor the position after disturbance, wherein the computing formula of b is:

$b=\mathrm{fit}\left(x_i\right)/\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\mathrm{fit}\left(x_i\right)$

Wherein fit (x _i) be the fitness value of i-th particle, N represents population size;

7): iterations s=s+1;

8): if iterations s meets s<iteration, upgrade population body position, jump to step 3); Otherwise terminate, now globally optimal solution is required defect profile.

Described step 2) in forward model be radial basis function neural network.

Described step 3) in hunting zone sharing policy be that all dimensions of single particle are reset in a certain specific solution space, global schema and local mode is divided into by the difference of Searching Resolution Space scope, under global schema, particle search scope is exactly the initial setting scope (x of particle _min, x _max); And under local mode, then from the current individual optimum solution Pbest of all particles, select maximal value Pbest _maxwith minimum value Pbest _min, composition (Pbest _min, Pbest _max), as the solution space that particle is new;

Separate the unicity that the original particle rapidity of the setting changing of sharing policy upgrades, computing formula is as follows:

$v_{i,j}(s+1)=\left\{\begin{array}{cc}{\mathrm{wv}}_{i,j}\left(s\right)+c_1{r}_{1i}\left(s\right)[{\mathrm{Pbest}}_{i,j}\left(s\right)-x_{i,j}\left(s\right)]+c_2{r}_{2i}\left(s\right)[{\mathrm{Gbest}}_j\left(s\right)-x_{i,j}\left(s\right)],& \mathrm{rand}\&GreaterEqual;{\mathrm{Ps}}_i\\ {\mathrm{wv}}_{i,j}\left(s\right)+c_1{r}_{1i}\left(s\right)[{\mathrm{Pbest}}_{i,j}\left(s\right)-x_{i,j}\left(s\right)]+c_2{r}_{2i}\left(s\right)[{\mathrm{Pbest}}_{a,j}\left(s\right)-x_{i,j}\left(s\right)],& \mathrm{rand}>{\mathrm{Ps}}_i\end{array}\right.$

Wherein w is inertia weight, and c is Studying factors, and r is the random number between 0 to 1, a is the particle sequence number randomly drawed from colony, and rand is the random number between 0 to 1, and Gbest is the globally optimal solution under current iteration, i and j is sequence number and the dimension of particle respectively, Ps _iformula be:

${\mathrm{Ps}}_i=0.5*\frac{(D-1)*\exp \left(\frac{i-1}{s-1}\right)-1}{2D}$

Wherein D is total dimension of particle.

Beneficial effect of the present invention is: the Magnetic Flux Leakage Inspecting Root cause analysis method that the present invention is based on the particle swarm optimization algorithm of improvement, the TSP question factor is introduced in EPUS-PSO algorithm, and applied to pipeline magnetic flux leakage defect reconstruct, to the defect of different size, defect profile can be reconstructed well by magnetic leakage signal.

Accompanying drawing explanation

Fig. 1 is existing iterative inversion frame principles schematic diagram figure;

Fig. 2 is Characterization of Real Defects Outlines in defect 1 situation of the present invention, reconstruct profile based on EPUS-PSO algorithm and the present invention compare schematic diagram;

Fig. 3 is Characterization of Real Defects Outlines in defect 2 situation of the present invention, reconstruct profile based on EPUS-PSO algorithm and the present invention compare schematic diagram;

Fig. 4 is Characterization of Real Defects Outlines in defect 3 situation of the present invention, reconstruct profile based on EPUS-PSO algorithm and the present invention compare schematic diagram;

Fig. 5 is Characterization of Real Defects Outlines in defect 4 situation of the present invention, reconstruct profile based on EPUS-PSO algorithm and the present invention compare schematic diagram;

Fig. 6 is schematic flow sheet of the present invention;

Fig. 7 is the experimental provision principle schematic that the embodiment of the present invention adopts;

Fig. 8 be the present invention in defect 5 situation Characterization of Real Defects Outlines, reconstruct the comparison diagram of profile based on EPUS-PSO algorithm and the present invention;

Fig. 9 be the present invention in defect 6 situation Characterization of Real Defects Outlines, reconstruct the comparison diagram of profile based on EPUS-PSO algorithm and the present invention.

Embodiment

Adaptive mutagenic factor adds in EPUS-PSO algorithm by the present invention, and is applied to pipeline magnetic flux leakage defect reconstructing method.To set forth from IEPUS-PSO algorithm technical scheme of the present invention below.

(1) IEPUS-PSO algorithm

Particle cluster algorithm is again flock of birds algorithm, is a kind of evolution algorithm proposed by doctor Eberhart and doctor kennedy by nineteen ninety-five, comes from the behavioral study to flock of birds predation.This algorithm is that the regularity being subject to flying bird cluster activity inspires at first, and then utilize swarm intelligence to set up simplified model.Particle cluster algorithm on animal cluster activity behavior observation basis, utilizes individual in population to make the motion of whole colony in solution space, produce evolutionary process from disorder to order to information shared, thus obtains optimum solution.

EPUS-PSO algorithm is a kind of improve PSO algorithm by effectively changing number of particles, and its maximum improvement is exactly the number effectively being changed particle by the fitness value change of the globally optimal solution of colony, and be called population management device, specific rules is:

If a the fitness value of () globally optimal solution does not all upgrade in double iteration, then increase a particle in colony, its position is:

$x_{\mathrm{add}}=\frac{\mathrm{Pbest}\left(a_1\right)+\mathrm{Pbest}\left(a_2\right)}{2}$

Wherein a ₁and a ₂the sequence number of two particles is randomly drawed in representative from current group, Pbest (a ₁) and Pbest (a ₂) be the current individual optimum solution of extracted two particles.If after adding this particle, when number of particles is greater than set maximum population, needs first to remove a particle that fitness value is the poorest, then add this new particle;

If b the fitness value of () globally optimal solution is all upgraded in double iteration, then that the poorest for fitness value particle enough, is then removed by the number that particle is described.

For prevent particle too early be absorbed in local optimum, other two large improvement are respectively hunting zone sharing policy reconciliation sharing policy, when activating probability P r (s) and being less than the random number of 0 to 1, then adopt hunting zone sharing policy, otherwise then adopt and separate sharing policy, the formula of Pr (s) is:

$\mathrm{Pr}\left(s\right)=0.03+\frac{0.07*s}{\mathrm{iteration}}$

In formula, iteration is the maximum times of iteration, and s is current iterations.

Hunting zone sharing policy is reset in a certain specific solution space all dimensions of single particle, global schema and local mode is divided into by the difference of Searching Resolution Space scope, under global schema, particle search scope is exactly the initial setting scope (x of particle _min, x _max).And under local mode, then from the current individual optimum solution Pbest of all particles, select maximal value Pbest _maxwith minimum value Pbest _min, composition (Pbest _min, Pbest _max), as the solution space that particle is new.

Separate the unicity that the original particle rapidity of the setting changing of sharing policy upgrades, computing formula is as follows:

$v_{i,j}(s+1)=\left\{\begin{array}{cc}{\mathrm{wv}}_{i,j}\left(s\right)+c_1{r}_{1i}\left(s\right)[{\mathrm{Pbest}}_{i,j}\left(s\right)-x_{i,j}\left(s\right)]+c_2{r}_{2i}\left(s\right)[{\mathrm{Gbest}}_j\left(s\right)-x_{i,j}\left(s\right)],& \mathrm{rand}\&GreaterEqual;{\mathrm{Ps}}_i\\ {\mathrm{wv}}_{i,j}\left(s\right)+c_1{r}_{1i}\left(s\right)[{\mathrm{Pbest}}_{i,j}\left(s\right)-x_{i,j}\left(s\right)]+c_2{r}_{2i}\left(s\right)[{\mathrm{Pbest}}_{a,j}\left(s\right)-x_{i,j}\left(s\right)],& \mathrm{rand}>{\mathrm{Ps}}_i\end{array}\right.$

Wherein w is inertia weight, and c is Studying factors, and r is the random number between 0 to 1, a is the particle sequence number randomly drawed from colony, and rand is the random number between 0 to 1, and Gbest is the globally optimal solution under current iteration, i and j is sequence number and the dimension of particle respectively, Ps _iformula be:

${\mathrm{Ps}}_i=0.5*\frac{(D-1)*\exp \left(\frac{i-1}{s-1}\right)-1}{2D}$

Wherein D is total dimension of particle.

The present invention, in order to improve the solving precision of total algorithm, adds a mutagenic factor in EPUS-PSO, carries out disturbance, constitute IEPUS-PSO algorithm to the position of all particles, and the formula of its mutagenic factor is: x' _i,j=x _i,j+ b*rand, x _i,jbe the original position of the jth dimension of i-th particle, i and j is sequence number and the dimension of particle respectively, and rand is the random number between 0 to 1, x' _i,jfor the position after disturbance, wherein the computing formula of b is:

$b=\mathrm{fit}\left(x_i\right)/\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{fit}\left(x_i\right)$

Wherein fit (x _i) be the fitness value of i-th particle, N represents population size;

In sum, the key step of IEPUS-PSO algorithm is as follows:

Step 101: the parameter arranging IEPUS-PSO algorithm, comprises maximum iteration time iteration, primary number and maximum number of particles, solution space dimension and solution space scope;

Step 102: set up fitness function, arranges iterations s=1;

Step 103: judge that particle adopts hunting zone sharing policy still to separate sharing policy;

Step 104: the fitness value calculating all particles;

Step 105: the current individual optimum solution and the globally optimal solution that upgrade particle colony;

Step 106: population management device adjusts population size, effectively changes the number of particle by the fitness value change of the globally optimal solution of colony;

Step 107: utilize mutagenic factor to carry out disturbance to the position of particle colony;

Step 108: iterations s=s+1;

Step 109: if iterations s meets s<iteration, then jump to step 103; Otherwise, terminate.

(2) the present invention is based on the pipeline magnetic flux leakage defect reconstructing method of IEPUS-PSO algorithm

According to the algorithm proposed in (1), be applied to the refutation process in pipeline magnetic flux leakage defect reconstructing method, reconstruction accuracy can be improved preferably.

Process flow diagram of the present invention as shown in Figure 1, particle position in algorithm represents the profile of defect, forward model adopts radial basis function neural network, the measurable magnetic leakage signal corresponding to profile, by prediction magnetic leakage signal compared with measured signal, calculate the error sum of squares between them, be defined as fitness function:

$F=\underset{d=1}{\overset{D}{\mathrm{\&Sigma;}}}{(p_d-y_d)}^2$

Wherein d is the dimension of particle, and D is total dimension of particle, p _dthe prediction magnetic leakage signal of forward model, y _dit is actual measurement magnetic leakage signal.Calculate the fitness value of particle, the more position of new particle, when iterations reaches maximal value, now the position of global optimum's particle is required defect profile.

(3) reconstitution experiments of the present invention and interpretation of result

The present invention uses emulation magnetic leakage signal and actual measurement magnetic leakage signal to verify validity of the present invention respectively, simulate signal is that finite element analysis software ANSYS emulates the 2 dimension defect profile-signal datas pair obtained, comprise 240 2 dimension defect sample, crack width is from 1 inch to 7 inches, and the degree of depth is not from 0.15 inch to 1.85 inches etc.240 sample centerings, 210 for training RBFNN, 30 are reconstructed defect for adopting the present invention.Actual measurement magnetic leakage signal by experiment device records.Because defect profile and magnetic leakage signal are 100 sampled points, so be also all 100 as the input layer of the radial basis function neural network of forward model and output layer nodes.The dispersion constant (spread) of neural network is 10 ^-8.

Table 1 gives comparing of the error sum of squares of defect profile and the actual profile adopting IEPUS-PSO algorithm and EPUS-PSO algorithm to reconstruct under different defect condition.

Table 1

One 2 dimension defect sample example is as shown in Fig. 2 to Fig. 5, and wherein solid line represents real profile, and dotted line represents the reconstruct profile of IEPUS-PSO algorithm, and dot-and-dash line is the reconstruct profile of EPUS-PSO algorithm.Upper as can be seen from figure, the algorithm after improvement can obtain defect profile more accurately.

In order to verify the performance of this patent institute extracting method further, actual measurement magnetic leakage signal is used to verify.The experimental provision that this example adopts as shown in Figure 7.

Experimental provision mainly comprises rotation platform, field coil, sensor, signal conditioning circuit, data collecting card, receiving terminal (PC) and motor.Defect is distributed in the edge surface of rotation platform.The yoke of excitation excitation is adopted to produce magnetic field.Yoke pole pitch rail level 1mm.Hall element probe is positioned at the centre position of yoke two magnetic pole of distance side 0.5mm, for obtaining magnetic leakage signal.After signal conditioning circuit regulates, magnetic leakage signal is sent to data collecting card.Finally, computing machine receives magnetic leakage signal.In addition, the speed of rotation platform is controlled by motor.

The material type of rotation platform upper surface is U71Mn.The defect distribution of different size is at the upper surface of rotation platform, and actual speed scope is 2 ~ 50m/s.The type of hall effect sensor and data collecting card is respectively UGN3503 and ADLINK DAQ 2204.Amplitude due to Analysis of Magnetic Flux Leakage Testing Signals is millivolt level, and the voltage range of data collecting card is volt level, therefore, adopts AD620 instrumentation amplifier to carry out the differential amplifier circuit that Design enlargement coefficient is 100.In addition, for avoiding pick-up unit remagnetization rotation platform, also design with the addition of demagnetizer.

Different from emulating the simulation magnetic leakage signal obtained, the true magnetic leakage signal that actual measurement obtains, owing to being collected by Hall element, comprises certain noise signal.Respectively the algorithm application improving front and back is carried out profile estimation to actual measurement magnetic leakage signal.Use iterations, error amount and be not originally put in table 2 computing time, Fig. 8 and Fig. 9 gives the result of reconstruct.Reconstructing method that this patent is carried as can be seen from Table 2, for survey magnetic leakage signal carry out profile reconstruct time, can when keep precision reduce computing time significantly.

Table 2

By the experiment of emulation magnetic leakage signal and the experiment of actual measurement magnetic leakage signal, can find out, IEPUS-PSO algorithm can improve reconstruction accuracy when computing time is substantially identical for emulation magnetic leakage signal, and for actual measurement magnetic leakage signal, when keeping reconstruction accuracy, computing time can be reduced significantly.

Claims (3)

1., based on a Magnetic Flux Leakage Inspecting Root cause analysis method for the particle swarm optimization algorithm improved, it is characterized in that, specifically comprise the steps:

1): the parameter that effective population Utilization strategies particle group optimizing IEPUS-PSO algorithm is set, comprise maximum iteration time iteration, primary number and maximum number of particles, solution space dimension and solution space scope, particle position represents the profile of defect;

2): set up fitness function wherein d is the dimension of particle, and D is total dimension of particle, p _dthe prediction magnetic leakage signal of forward model, y _dbe actual measurement magnetic leakage signal, iterations s=1 is set;

3): judge that particle adopts hunting zone sharing policy still to separate sharing policy, when activating probability P r (s) and being less than the random number of 0 to 1, then adopt hunting zone sharing policy, otherwise then adopt solution sharing policy, the formula of Pr (s) is:

$\mathrm{Pr}\left(s\right)=0.03+\frac{0.07*s}{\mathrm{interation}}$

Wherein iteration is maximum iteration time, and s is current iteration number of times;

: the fitness value calculating all particles according to fitness function, and upgrade current individual optimum solution and the globally optimal solution of particle colony 4);

5): population management device adjusts population size according to IEPUS-PSO algorithm, effectively changed the number of particle by the fitness value change of the globally optimal solution of colony, specific rules is:

If a the fitness value of () globally optimal solution does not all upgrade in double iteration, then increase a particle in colony, its place value is:

$x_{\mathrm{add}}=\frac{\mathrm{Pbest}\left(a_1\right)+\mathrm{Pbest}\left(a_2\right)}{2}$

Wherein a ₁and a ₂the sequence number of two particles is randomly drawed in representative from current group, Pbest (a ₁) and Pbest (a ₂) be the current individual optimum solution of extracted two particles, if after adding this particle, when number of particles is greater than set maximum population, needs first to remove a particle that fitness value is the poorest, then add this new particle;

If b the fitness value of () globally optimal solution is all upgraded in double iteration, then that the poorest for fitness value particle enough, is then removed by the number that particle is described;

6): utilize mutagenic factor to carry out disturbance to the position of all particles, the formula of mutagenic factor is:

X' _i,j=x _i,j+ b*rand, x _i,jbe the original position of the jth dimension of i-th particle, i and j is sequence number and the dimension of particle respectively, and rand is the random number between 0 to 1, x' _i,jfor the position after disturbance, wherein the computing formula of b is:

$b=\mathrm{fit}\left(x_i\right)/\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{fit}\left(x_i\right)$

Wherein fit (x _i) be the fitness value of i-th particle, N represents population size;

7): iterations s=s+1;

8): if iterations s meets s<iteration, upgrade population body position, jump to step 3); Otherwise terminate, now globally optimal solution is required defect profile.

2., according to claim 1 based on the Magnetic Flux Leakage Inspecting Root cause analysis method of particle swarm optimization algorithm improved, it is characterized in that, described step 2) in forward model be radial basis function neural network.

3. according to claim 1 based on the Magnetic Flux Leakage Inspecting Root cause analysis method of the particle swarm optimization algorithm improved, it is characterized in that, described step 3) in hunting zone sharing policy be that all dimensions of single particle are reset in a certain specific solution space, global schema and local mode is divided into by the difference of Searching Resolution Space scope, under global schema, particle search scope is exactly the initial setting scope (x of particle _min, x _max); And under local mode, then from the current individual optimum solution Pbest of all particles, select maximal value Pbest _maxwith minimum value Pbest _min, composition (Pbest _min, Pbest _max), as the solution space that particle is new;

Separate the unicity that the original particle rapidity of the setting changing of sharing policy upgrades, computing formula is as follows:

$v_{i,j}(s+1)=\left\{\begin{array}{cc}{\mathrm{wv}}_{i,j}\left(s\right)+c_1{r}_{1i}\left(s\right)[{\mathrm{Pbest}}_{i,j}\left(s\right)-x_{i,j}\left(s\right)]+c_2{r}_{2i}\left(s\right)[{\mathrm{Gbest}}_j\left(s\right)-x_{i,j}\left(s\right)],& \mathrm{rand}\&GreaterEqual;{\mathrm{Ps}}_i\\ {\mathrm{wv}}_{i,j}\left(s\right)+c_1{r}_{1i}\left(s\right)[{\mathrm{Pbest}}_{i,j}\left(s\right)-x_{i,j}\left(s\right)]+c_2{r}_{2i}\left(s\right)[{\mathrm{Pbest}}_{a,j}\left(s\right)-x_{i,j}\left(s\right)],& \mathrm{rand}>{\mathrm{Ps}}_i\end{array}\right.$

Wherein w is inertia weight, and c is Studying factors, and r is the random number between 0 to 1, a is the particle sequence number randomly drawed from colony, and rand is the random number between 0 to 1, and Gbest is the globally optimal solution under current iteration, i and j is sequence number and the dimension of particle respectively, Ps _iformula be:

${\mathrm{Ps}}_i=0.5*\frac{(D-1)*\exp \left(\frac{i-1}{s-1}\right)-1}{2D}$

Wherein D is total dimension of particle.