CN115795341A - Two-dimensional piston pump health state assessment method based on variable rotating speed - Google Patents
Two-dimensional piston pump health state assessment method based on variable rotating speed Download PDFInfo
- Publication number
- CN115795341A CN115795341A CN202211400126.9A CN202211400126A CN115795341A CN 115795341 A CN115795341 A CN 115795341A CN 202211400126 A CN202211400126 A CN 202211400126A CN 115795341 A CN115795341 A CN 115795341A
- Authority
- CN
- China
- Prior art keywords
- frequency
- piston pump
- svm
- data
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention provides a two-dimensional piston pump health state evaluation method based on variable rotating speed, which comprises the following steps: data acquisition: collecting vibration signals reflecting the normal state and the fault state of the pump, and preprocessing data: the method comprises the following steps of (1) extracting instantaneous frequency by adopting GLCT (global level-matching pursuit) to perform equiangular resampling to eliminate the influence of rotating speed, then performing order analysis on data to obtain order domain characteristic parameters, constructing a time-frequency fault sample set by using a time-frequency analysis method, and training and optimizing a model: and constructing a fault diagnosis model, taking the training sample set as input and the state label as output, and training the model. The vibration signal of the two-dimensional (2D) piston pump is input into the trained and optimized fault diagnosis model, the fault diagnosis model outputs the state label value, the final health state classification recognition result can be directly obtained, the method is simple and reliable, and the accuracy of the diagnosis result is ensured.
Description
Technical Field
The invention relates to the technical field of hydraulic element health state assessment, in particular to a two-dimensional piston pump health state assessment method based on variable rotating speed.
Background
The microminiature two-dimensional piston pump is a hydraulic pump with a novel structure, forms a novel flow distribution mode by utilizing the motion principle of two degrees of freedom of a piston, has the characteristic of realizing high power-weight ratio and simultaneously considering microminiaturization of the structure, and has the advantages of simple structure, small volume, light weight, few friction pairs and quick response compared with the traditional hydraulic pump. The health state of the robot is closely related to the reliability and stability of the overall operation of the equipment. Therefore, the method can accurately evaluate the health state of the two-dimensional piston pump, namely the 2D piston pump, can effectively solve the problem that the hydraulic system is difficult to find out faults, and has great practical significance for formulating a reasonable maintenance plan and guaranteeing the stability and the safety of equipment.
With the rapid development of signal processing technology and artificial intelligence, the artificial intelligence is applied to the evaluation and analysis of the health state of the two-dimensional piston pump, and a new idea is provided for realizing the real-time and accurate evaluation of the current health state of the system, predicting the performance degradation degree and development trend of the system, finding early faults and realizing state-based visual maintenance.
The two-dimensional piston pump health status recognition is actually a process of classifying and recognizing according to the characteristics of vibration signal reaction faults and damage degrees, and the vibration signals collected according to the damage degrees of the test objects theoretically have vibration characteristics corresponding to the damage degrees. Therefore, data processing and analysis are carried out according to the collected vibration signals, and health state research based on data driving is a hot topic for health diagnosis and identification of mechanical equipment. However, at present, most of the researches on the steady-state constant rotating speed working condition are conducted at home and abroad, and the researches on the variable rotating speed working condition are relatively rare.
The vibration response of the variable-speed working condition is very complex in time domain, frequency domain and time-frequency domain, and has large influence on the essential characteristics of signals. In order to eliminate the influence of speed variation, a mapping relation between the rotating speed and the vibration signal is found by researching the relation between the rotating speed and the vibration signal, and the method is an idea for eliminating the influence of the rotating speed. One important method for identifying the variable-speed health state is order analysis, and two methods are mainly used for common order analysis: hardware order analysis method, order analysis method based on instantaneous frequency estimation. Hardware order analysis requires synchronous sampling by installing hardware, so a phase reference signal acquired from an encoder or a tachometer is necessary, but many devices are not always capable of installing encoders and tachometers due to budget cost or technical reasons. In some important devices, such as an aircraft and a robot, the sensor is inconvenient to install. Therefore, the instant frequency of the vibration signal is accurately estimated through a parameterized time-frequency analysis method, so that the classification and identification of the fault are carried out, which is the core of the order analysis method of the instant frequency estimation.
The influence of the change of the rotating speed of the obtained vibration signal on the vibration signal is eliminated, effective information in the signal is mined, a complete fault classification system is established, and the classification accuracy of the health state of the two-dimensional piston pump is greatly influenced, so that the research on the classification method is very critical to obtain the health state of the two-dimensional piston pump. With the increasingly optimized development of classification algorithms, the classification methods are endless, and the selection of a suitable classification algorithm for a specific study object is the current research direction. The two-dimensional piston pump health state identification is a small sample data classification test, so that the two-dimensional piston pump health state identification accuracy can be higher than that of the two-dimensional piston pump health state identification by adopting a support vector machine to classify the small sample data and carrying out parameter optimization.
Disclosure of Invention
In order to solve the defects of the prior art, the invention provides a two-dimensional piston pump health state assessment method based on variable rotating speed, which can monitor and identify the health state of a two-dimensional piston pump based on variable rotating speed and intelligently identify the health state of the two-dimensional piston pump with high precision. Compared with other traditional methods CPA-SVM, the evaluation method has higher classification accuracy, multiple solutions are initialized and iterative enhancement is carried out on the basis of the CPA of the population algorithm, and although the algorithm needs more function evaluation, local optimization can be avoided due to information sharing, so that the optimization of a target function can be effectively carried out, and an optimal health monitoring model is ensured to be obtained.
Specifically, the invention provides a two-dimensional piston pump health state assessment method based on variable rotating speed, which comprises the following steps:
s1, data acquisition: collecting vibration signals reflecting normal states and fault states of the two-dimensional piston pump as original state data;
s2, data preprocessing and analyzing: carrying out GLCT time-frequency analysis on the original state data of the two-dimensional piston pump vibration signal acquired in the step S1, and extracting the instantaneous rotating speed frequency of the vibration signal;
s3, extracting angular domain vibration signals and order domain characteristic parameters, wherein the method comprises the following substeps:
s31, performing data fitting processing on the instantaneous rotating speed frequency extracted in the step S21, intercepting data in a frequency rising stage of the instantaneous frequency by adopting a segmented multi-order polynomial fitting method to perform data fitting to obtain a data fitting curve, and obtaining an order spectrum according to the fitting curve;
s32, for the angular domain vibration signal S (L) of the two-dimensional piston pump, if the length of the data sampled at equal angular intervals is L, the peak index Cf, the pulse index If and the kurtosis index K of the angular domain vibration signal S (L) are V The calculation formulas of (A) are respectively as follows:
wherein S is rms Is the root mean square value, S max In the form of a peak value, the peak value,
is the absolute average amplitude, β is the kurtosis;
root mean square value S rms The calculation formula of (2) is as follows:
peak value S max The calculation formula of (2) is as follows:
S max ＝max(|s(l)|)
absolute average amplitude
The calculation formula of (2) is as follows:
the kurtosis β is calculated as:
let the angular domain vibration signal s (L) order spectrum be Sl (m), where L =1,2, …, L, m is the order data length variable, and take m =1,2, …, D max In which D is max For maximum order, define the RMS value Y of the order spectrum 1 The following were used:
and use of Y 1 Size measurement ofThe vibrational energy intensity of the entire order spectrum;
s33, analyzing by using the amplitudes corresponding to the 11, 22, 33 and 44 orders of the order spectrum as order domain characteristic parameters:
wherein n is the number of the interval order spectrum data, Y 2 、Y 3 、Y 4 、Y 5 The amplitudes at orders 11, 22, 33 and 44 of the order spectrum, respectively;
s4, training and optimizing a fault diagnosis model, and specifically comprises the following substeps:
s41, constructing a fault diagnosis model: mixing Cf, if and K V 、Y 1 、Y 2 、Y 3 、Y 4 、Y 5 Constructing a fault diagnosis model as a training characteristic parameter of a CPA-SVM optimization classification algorithm, dividing a sample library into a training sample set and a test sample set according to a proportion, taking the training sample set as the input of the fault diagnosis model, and taking a state label as the output of the fault diagnosis model;
s42, accurately optimizing model parameters by using a CPA (cross correlation algorithm) optimization algorithm according to a punishment factor c and a radial cardinal number g in a large range under a target function of the classification accuracy of the SVM (support vector machine);
s43, performing fault classification model training by using the optimized CPA-SVM to obtain optimal model parameters best c and best g so as to obtain a trained and optimized fault diagnosis model;
s5, model verification: calling the trained and optimized fault diagnosis model, using the test sample set as the input of the trained and optimized fault diagnosis model, using the state label as the output of the trained and optimized fault diagnosis model, and verifying the comprehensive performance of the diagnosis model;
s6, model diagnosis: inputting the vibration signal of the two-dimensional piston pump into the trained and optimized fault diagnosis model, and outputting a state label value by the fault diagnosis model to obtain a health state classification and identification result.
Preferably, step S2 specifically comprises the following sub-steps:
step S21, constructing an original STFT formula as follows:
wherein w (u-t') is a certain window;
step S22, on the basis of the STFT formula, a series of discrete demodulation operators are used for approximating the optimal demodulation operator
The STFT formula based on the discrete demodulation operator is obtained as follows:
step S23, for each TF point (t', w), if the demodulation operator is discrete
Close to the modulation component of the signal, TF around its IF indicates a high energy concentration, whose amplitude | S (t ', w, c) | reaches a maximum value among all values, after which, for each TF point, the optimal parameter c ' of the parameter c is obtained from the amplitude of | S (t ', w, c) |:
the time-frequency representation of the proposed time-frequency analysis method is as follows:
GS(t',w)＝S(t',w,c')
the formula of the frequency spectrum calculation is as follows:
Spec(t',w)＝|GS(t',w)| 2
step S24, determining discrete demodulation operator
Introduction screwAnd (3) converting a parameter a to realize rotation on a time-frequency plane:
wherein, T s For sampling, F s Is the sampling frequency;
setting N values of the rotation parameter a, and dividing a time frequency plane into N +1 sections in an average way:
a＝-π/2+π/(N+1),-π/2+2·π/(N+1)...-π/2+N·π/(N+1)；
step S25, ensuring demodulation operator
All feasible modulation components in the signal can be described, and the STFT formula of the discrete demodulation operator is rewritten to obtain a GLCT formula:
and S26, carrying out GLCT time-frequency analysis according to the GLCT formula in the step S25, and extracting the instantaneous rotating speed frequency S (t', w, a) of the vibration signal.
Preferably, step S31 specifically comprises the following sub-steps:
s311, expressing the relation between the rotating speed and the first-order instantaneous frequency as follows:
wherein f (t) is the shaft rotation frequency, and P (t) is the reference shaft rotation speed;
s312, obtaining a fitting curve by adopting a piecewise multi-order polynomial fitting method:
R k (t)＝a k +b k t+c k t 2
wherein Rk (t) is a function of each section of rotating speed curve, k is the serial number of the section, a k 、b k 、c k Is a polynomial coefficient;
s313, carrying out integral equation solving on the rotating speed fitting curve to obtain an equiangular interval interpolation phase discrimination time scale time sequence, wherein the solving process is as follows:
wherein, T n Is a key phase time scale, n is a time scale serial number, T 0 Is the initial fitting time; substituting the formula into a fitting curve formula, and obtaining a group of equiangular interval interpolation phase discrimination time scale time sequence T through effective algebraic solution n ：
S314, based on the equiangular interval interpolation phase discrimination time scale time sequence obtained in the step S313, smoothing is completed by utilizing a Lagrange value method to obtain an angle domain signal;
and S315, carrying out Fourier transform on the angle domain signal data obtained in the step S314 to obtain an order spectrum.
Preferably, 200 sets of data are acquired and divided into a training sample set and a testing sample set in a proportion of 4:1, and the number of iterations in the set group _ iter =20, the attraction rate absorption _ rate =0.8, the growth rate growth _ rate =2, the reproduction rate reproduction _ rate =1.8, the number of carnivorous plants ncplan =10, and the number of prey =20 are defined.
Preferably, step S42 is specifically: 8 feature vectors obtained through time-frequency domain and order analysis are used as training feature parameters of a CPA-SVM optimization classification algorithm, target labels 1,2, 3 and 4 in the last column are four different wear states corresponding to the feature vectors according to the optimization classification algorithm, then the ranges of c and g are determined, and the c and g in the large range are accurately optimized by the CPA optimization algorithm under the objective function of the classification accuracy of the SVM.
Preferably, in step S1, the vibration signal is acquired by a vibration sensor mounted on the two-dimensional piston pump.
Preferably, the optimal parameters are best c =224.6036, best g =31.5104.
Preferably, the step S42 includes the following steps:
the meat eating algorithm searches for a proper SVM parameter value in continuous iterative evolution, so that the accuracy of SVM classification is improved, the mathematical model can be simplified to search for an optimal decision vector [ c, g ], wherein c and g respectively represent a punishment parameter vector and a kernel function vector of the SVM, and the following aims are achieved by maximization:
max F＝SVM_Acc
wherein F represents an objective function, SVM _ Acc represents the classification accuracy of SVM, and M T Indicating the number of samples correctly classified into the corresponding class, M F Indicating the number of categories which are not correctly classified into corresponding categories; firstly, defining the iteration times group _ iter =20, the attraction rate _ rate =0.8, the growth rate growth _ rate =2, the reproduction rate _ rate =1.8, the number of carnivorous plants nCPLant =10 and the number of preys nPrey =20 in a group, in an objective function SVM _ Acc, the maximum iteration times T =40, firstly, roughly optimizing parameters by a grid search method for parameters c and g of an optimization algorithm SVM, and determining the ranges of c and g to be 2 respectively -2 -2 8 I.e. 0.25-256 and 2 -5 -2 5 I.e. 0.03-32.
Compared with the prior art, the invention has the following effects:
(1) The invention takes the original vibration signal of the two-dimensional (2D) piston pump as a data source, can carry out nondestructive monitoring on the piston pump, adopts a keyless phase order analysis method to eliminate the influence of the rotating speed of the two-dimensional (2D) piston pump, and has more accurate monitoring result.
(2) Because the traditional instantaneous frequency extraction method has the defects of low time-frequency resolution and the like, so that the instantaneous frequency error is larger, the method can more accurately extract the instantaneous frequency by adopting a GLCT time-frequency analysis method, and then carry out angular domain resampling to further finish the order analysis in the angular domain.
(3) The two-dimensional (2D) piston pump health state evaluation method based on the vibration signals utilizes dimensionless parameters insensitive to speed change in the vibration signals in the angle domain, and can accurately reflect the faults of the two-dimensional (2D) piston pump, such as the peak index, the pulse index and the kurtosis index of the vibration signals in the angle domain. Aiming at the aspect of order domain, order energy and the like are found to be obvious to faults, and a high-dimensional and complex overall fault sample set can be constructed by using the root mean square value of an order spectrum and the order amplitude of the order domain as characteristic parameters.
(4) The two-dimensional (2D) piston pump health state evaluation method based on the vibration signal fully utilizes a CPA method to optimize the two parameters, and iteratively searches parameter penalty factors c and radial cardinal numbers g of SVM multi-class prediction with high accuracy for multiple times under the judgment of a fitness objective function. The two-dimensional (2D) piston pump health status recognition capability is improved.
(5) The two-dimensional (2D) piston pump health state evaluation method based on the vibration signal is based on training and testing of characteristic parameters of an angle domain and an order domain, and can successfully solve the problems of high-dimensional design variables, existence of various constraints, a search space with a plurality of local optimal solutions and the like after the SVM is optimized through CPA, so that the capability of identifying the two-dimensional (2D) piston pump health state is improved, and the accuracy of model output is ensured.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is a flow chart of the implementation of the method for identifying the health state of the two-dimensional piston pump based on the vibration signal according to the present invention;
fig. 3 is a comparison graph of instantaneous frequencies of vibration signals of four states of a two-dimensional piston pump measured actually according to an embodiment of the present invention, in which fig. 3a is a schematic diagram of a healthy state, fig. 3b is a schematic diagram of a light wear failure, fig. 3c is a schematic diagram of a medium wear failure, and fig. 3d is a schematic diagram of a heavy wear failure;
fig. 4 is an angle domain resampling comparison diagram of vibration signals of four states of a two-dimensional piston pump measured actually in the embodiment of the present invention, where fig. 4a is a schematic diagram of a healthy state, fig. 4b is a schematic diagram of a light wear fault, fig. 4c is a schematic diagram of a medium wear fault, and fig. 4d is a schematic diagram of a heavy wear fault;
FIG. 5 is a schematic diagram of a fault classification model for SVM optimization by CPA-SVM optimization algorithm according to the embodiment of the present invention, wherein the optimal fitness is 93.5%;
fig. 6 is a two-dimensional piston pump health status classification result graph based on SVM according to the embodiment of the present invention, in which fig. 6a is a health status diagnosis result graph, and fig. 6b is a diagnosis result confusion matrix graph.
Detailed Description
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
The invention will be further described with reference to the following figures and specific examples, but the scope of the invention is not limited thereto. Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
Under the working condition of variable rotating speed, the change of the rotating speed can cover sensitive information of related degradation indexes, and the change of vibration signals in time domain and frequency domain is very complex and severe, so that the conventional method for positioning the health state of the piston pump through sensitive characteristics is not suitable. Therefore, how to eliminate the influence of variable rotating speed and position the degradation state index of the piston pump is an important problem of the invention. According to the method, a non-constant-rotating-speed piston pump abrasion fault test bed is built, data collection is carried out on piston pumps with different abrasion degrees, a coder-free test method is adopted, instantaneous frequency is extracted based on GLCT to carry out equal-angle resampling to eliminate the influence of rotating speed, and then data are subjected to order analysis to explore the health state of the piston pumps.
Specifically, the two-dimensional piston pump with different piston wear degrees is subjected to data acquisition through the vibration sensor arranged on the two-dimensional piston pump, vibration signals reflecting the normal state and the health state evaluation state of the pump are acquired and serve as original state data, then the instantaneous frequency of the strong nonlinear signal can be accurately extracted by adopting a GLCT (global level-shift keying) method, the rotating speed corresponding to the rotating frequency is obtained from the mapping of the instantaneous frequency and the instantaneous rotating speed, and a foundation is laid for further equal-angle domain resampling and variable rotating speed influence elimination. The time-frequency analysis method of GLCT can more accurately extract the instantaneous frequency, then carry out angle domain resampling, carry out Fourier transform on the data of the angle domain resampling, and obtain an order spectrum to carry out order analysis on the data; then extracting some dimensionless indexes of an angle domain which is basically irrelevant to the operation condition and insensitive to the speed change as time domain characteristic parameters and order spectrum root mean square values as order domain characteristic parameters as total characteristic parameters; and finally, classifying the characteristic parameters by adopting a one-to-one method in Support Vector Machine (SVM) multi-classification, constructing a training model, then optimizing a radial basis kernel parameter g and a support vector machine punishment factor c, which influence the sample training and testing process by the value size, by adopting a Carnivorous Plant Algorithm (CPA), and establishing a CPA-SVM health state assessment classification training model to shorten the time for establishing the health state assessment classification model and improve the accuracy of SVM classification. And analyzing the surface of the test result, and comparing the obtained test result with classification accuracy of the SVM and ELM methods. The result shows that the health state assessment and recognition accuracy rate of the CPA-SVM model optimized based on the carnivorous plant algorithm is 93.75%, compared with the SVM model which is not optimized, the accuracy rate is increased by 11.25%, meanwhile, the classification accuracy rate is also higher than ELM, and the advantage of the carnivorous plant algorithm optimization support vector machine method for health state recognition is reflected.
The specific steps of the research on the method for evaluating the health state of the variable-speed two-dimensional piston pump are as follows:
the invention provides a CPA-SVM-based variable-rotation-speed two-dimensional piston pump health state evaluation method, which mainly comprises the following two steps: firstly, extracting the instantaneous frequency of a vibration signal by adopting a time-frequency analysis method based on GLCT, carrying out angle domain resampling, carrying out Fourier transform on the data subjected to angle domain resampling, and carrying out order analysis on the data by obtaining an order spectrum. And then, four hundred groups of data corresponding to four health state evaluation states are respectively established according to the obtained characteristic parameters, and health state evaluation classification model training is carried out by a CPA-SVM-based classification optimization method, so that the advantage of CPA-SVM in two-dimensional piston pump health state recognition is embodied.
As shown in fig. 1 and fig. 2, it specifically includes the following steps:
s1, different health states of the two-dimensional piston pump are reflected by replacing pistons with different wear degrees, vibration signals of the two-dimensional piston pump with four different wear degrees are acquired respectively, the test sampling frequency is 20000hz, data processing and down-sampling are 5120hz, and the sampling time is 35S. Four health status assessments 100 sets of data were collected for each, for a total of 400 sets of data.
S2, carrying out GLCT time-frequency analysis on the 4 x 100 groups of collected data of the four groups of variable-rotation-speed two-position piston pump vibration signals, and extracting instantaneous rotation speed frequency. The method for extracting the instantaneous frequency by adopting the time-frequency analysis method based on the GLCT comprises the following steps:
approximating optimal demodulation operators using a series of discrete demodulation operators
The STFT formula for considering the discrete demodulation operator is given as follows:
for each TF point (t', w) if the demodulation operator is discrete
Close to the modulation component of the signal, TF around its IF indicates a higher energy concentration, its amplitude | S (t', w, c) | reaches a maximum value among all values. Then, for each TF point, according to the amplitude of | S (t', w, c) |, an optimal parameter c is obtained:
the time-frequency representation of the proposed time-frequency analysis method is as follows:
GS(t',w)＝S(t',w,c')
the spectrum can be defined as:
Spec(t',w)＝|GS(t',w)| 2
next, a discrete demodulation operator is determined
According to the existing research results, by introducing
Will have a rotating effect on the time frequency result, with a degree of rotation arctan (-c). For a signal, if the sampling time (Ts) and the sampling frequency (Fs) are determined, a time-frequency plane of t ∈ (0, ts), f ∈ (0, fs/2) is determined. Therefore, by introducing a rotation parameter a, the rotation on the time-frequency plane is realized, such as:
ensuring demodulation operator
Being able to describe all possible modulation components in the signal, the above formula can be rewritten as:
the parameter a is set to have N values, and the time frequency plane can be divided into N +1 sections in an average way:
a＝-π/2+π/(N+1),-π/2+2·π/(N+1)...-π/2+N·π/(N+1)
as can be seen from the above equation, the method GLCT used in the present invention introduces a parameter a more than STFT.
And S3, carrying out angle domain resampling on the tested equipment without rotating speed estimation to obtain instantaneous frequency, carrying out data fitting processing on the obtained instantaneous frequency, and particularly adopting a segmented multi-order polynomial fitting method to intercept the frequency rising stage of the instantaneous frequency to carry out data fitting to obtain a fitting curve.
For the device to be tested under the condition of no rotating speed estimation to carry out resampling, the calculation of key phase time scales is needed. In order to perform data fitting processing on the obtained instantaneous frequency, the chapter intercepts the frequency rising stage of the instantaneous frequency to perform data fitting, firstly obtains a rotating speed curve, and the relationship between the rotating speed and the first-order instantaneous frequency is as follows:
where f (t) is the shaft rotation frequency and P (t) is the reference shaft rotation speed.
And converting the discrete data into a smooth curve by a fitting method through the obtained discrete rotating speed information, and then further simulating the rotating speed to obtain a model. The high-precision fitting effect is achieved by adopting a segmented multi-order polynomial fitting method:
R k (t)＝a k +b k t+c k t 2
wherein Rk (t) is a function of each section of rotating speed curve, k is the serial number of the section, a k 、b k 、c k Is a polynomial coefficient. The result obtained by solving the integral equation of the rotating speed fitting curve is a phase discrimination time scale sequence, and the solving process is as follows:
wherein T is n For key phase time scale, n is time scale sequence number, T 0 Is the initial fitting time. Substituting the formula into the previous calculation formula of Rk (T), and obtaining a group of equiangular interval interpolation phase discrimination time scale time sequence T by obtaining an effective algebraic solution n ：
The fitting curve and the angle resampling data obtained by the method are used for carrying out Fourier transform on the data after the angular domain resampling, so that the problem of spectrum ambiguity of directly carrying out Fourier transform on the collected vibration signals can be solved, and the order spectrum is obtained.
For an angular domain vibration signal s (L) of a two-dimensional piston pump, if the length of data sampled at equal angle intervals is L, a peak index Cf, a pulse index If and a kurtosis index K are obtained V Are respectively defined as follows:
wherein S rms Root mean square value, defined as:
peak value S max Is defined as:
S max ＝max(|s(l)|)
the absolute average amplitude is defined as:
kurtosis β is defined as:
let the angular domain vibration signal s (L) (L =1,2, …, L) order spectrum be Sl (m), where m represents the order data length variable, take m =1,2, …, D max And defining the RMS value of the order spectrum if Dmax is the maximum order:
and use of Y 1 The magnitude is a measure of the vibration energy intensity of the entire order spectrum.
In order to better represent the characteristic differences under different wear degrees, the amplitude values corresponding to the 11, 22, 33 and 44 orders of the order spectrum are used as characteristic parameters for analysis, because in the process of calculating the order spectrum, the maximum amplitude value cannot be exactly appeared on the integer order of each signal, so the average value of the 11, 22, 33 and 44 order intervals +/-0.15 is selected as the characteristic value to complete the selection, namely:
in the formula, n is the number of the order spectrum data in the interval.
S4, selecting a peak index Cf, a pulse index If and a kurtosis index KV in dimensionless indexes as degradation characteristic parameters in the aspect of an angle domain, and marking as X = [ Cf, if, K = V ]Selecting the average value of 11, 22, 33 and 44 order intervals +/-0.15 as the health state evaluation characteristic parameter of the order domain, and recording as Y = [ Y = 1 ,Y 2 ,Y 3 ,Y 4 ,Y 5 ]And the two are combined into a total characteristic parameter which is recorded as Z = [ Cf, if, K = V ,Y 1 ,Y 2 ,Y 3 ,Y 4 ,Y 5 ]。
And S5, mapping the health state evaluation characteristic parameter sample to a higher-dimensional space by adopting a Gaussian radial basis function. The quadratic optimization problem is solved, the value of the normal vector w intercept b on the hyperplane is obtained by using the mathematical expression of the optimal hyperplane, and the optimization problem is converted into a dual problem by using a Lagrange multiplier method to obtain an optimal classification function.
And S6, training 80 of 100 groups of data and testing the remaining 20 groups. And then roughly optimizing the parameters by a grid search method to determine the ranges of c and g. And under the objective function of the classification accuracy of the SVM, accurately optimizing the parameters of c and g in a large range by using a CPA optimization algorithm. And further performing health state evaluation classification model training by using the CPA-SVM, wherein the obtained optimal parameters are best c =224.6036 and best g =31.5104, the objective function of the model is best scored, and the rest parameters are default values.
The specific method comprises the following steps: the first 80 of the 4 x 100 groups of signals, i.e., a total of 320 groups of signals, were trained separately, and the remaining 4 x 20 groups, i.e., a total of 80 groups, were tested separately. C f, if, K V 、Y 1 、Y 2 、Y 3 、Y 4 、Y 5 As training characteristic parameters of a CPA-SVM optimization classification algorithm, target labels 1,2, 3 and 4 in the last column are four different wear states corresponding to the optimization classification algorithm according to characteristic vectors, and then according to a determined penalty factor c and the range of a radial base number g, under the objective function of the classification accuracy of the SVM, c and g in a large range are used for accurately optimizing the parameters by using the CPA optimization algorithm. And further performing health state evaluation classification model training by using the CPA-SVM, wherein the obtained optimal parameters are best c =224.6036 and best g =31.5104, the objective function of the model is best scored, and the rest parameters are default values.
The specific embodiment is as follows:
a two-dimensional piston pump test bed is built, different health state evaluation states of a two-dimensional (2D) piston pump are simulated by replacing pistons with different abrasion degrees in the test, vibration signals are acquired for the two-dimensional (2D) piston pump, and finally the signals are processed by using the method provided by the invention. The test sorts out pistons of different degrees of wear from two-dimensional (2D) piston pumps evaluated for state of health, in order of state of health, light wear, moderate wear and heavy wear. As shown in fig. 3a to 3d, fig. 3a is a schematic diagram of the instantaneous frequency of the vibration signal in the healthy state, fig. 3b is a schematic diagram of the instantaneous frequency of the vibration signal in the light wear state, fig. 3c is a schematic diagram of the instantaneous frequency of the vibration signal in the medium wear state, and fig. 3d is a schematic diagram of the distribution of the instantaneous frequency of the vibration signal in the heavy wear state. As can be seen from fig. 3a to 3D, it is difficult to determine the different health states of the two-dimensional (2D) piston pump corresponding to the pressure signals by observing the difference of the instantaneous waveform diagrams.
And actually measuring a vibration signal angle frequency domain diagram under the health state extracted by a two-dimensional (2D) piston pump generalized linear frequency modulation GLCT. As shown in fig. 4 a-4 d, fig. 4a is a healthy state vibration signal angle frequency domain diagram, fig. 4b is a light wear vibration signal angle frequency domain diagram, fig. 4c is a medium wear vibration signal angle frequency domain diagram, and fig. 4d is a heavy wear health state evaluation vibration signal angle frequency domain diagram.
Four wear states collected by the test bed are all carried out under the same working condition, the pulse signal of the stepping motor passes through two stages, the first stage is an ascending stage that the pulse signal is increased from 5000 to 15000 in 500 per second, the second stage is a descending stage that the pulse signal is decreased from 15000 to 2500 in 1000 pulses, and the whole signal collection time is 32 seconds from 1 second to 33 seconds. The known stepping motor has a fine value of 8, and the change of the rotating speed can be converted into 200r/min to 600r/min.
And forming 8 eigenvectors by using the peak index, the pulse index and the kurtosis index extracted from the time-frequency domain and 11, 22, 33 and 44 order data extracted from the order domain to identify the health state of the two-dimensional (2D) piston pump at variable rotation speed. Out of the 200 sets of data, 180 sets were trained and the remaining 20 sets were tested. The number of iterations in the group _ iter =20, attraction rate adherence _ rate =0.8, growth rate growth _ rate =2, reproduction rate _ rate =1.8, number of carnivorous plants ncplan =10 and number of prey =20 are defined.
8 feature vectors obtained through time-frequency domain and order analysis are used as training feature parameters of a CPA-SVM optimization classification algorithm, target labels 1,2, 3 and 4 in the last column are four different wear states corresponding to the feature vectors according to the optimization classification algorithm, then the ranges of a penalty factor c and a radial base number g are determined, and under the target function of the classification accuracy of the SVM, the c and g in a large range are accurately optimized through the CPA optimization algorithm. And further performing health state evaluation classification model training by using the CPA-SVM, wherein the obtained optimal parameters are best c =224.6036 and best g =31.5104, and the objective function score of the model is best.
The optimization process specifically comprises the following steps:
the step S42 specifically includes:
the meat eating algorithm searches for a proper SVM parameter value in continuous iterative evolution, so that the accuracy of SVM classification is improved, the mathematical model can be simplified to search for an optimal decision vector [ c, g ], wherein c and g respectively represent a punishment parameter vector and a kernel function vector of the SVM, and the following aims are achieved by maximization:
max F＝SVM_Acc
wherein F represents an objective function, SVM _ Acc represents the classification accuracy of SVM, and M T Indicating the number of samples correctly classified into the corresponding class, M F Indicating the number of categories which are not correctly classified into corresponding categories; firstly, defining the iteration times group _ iter =20, the attraction rate _ rate =0.8, the growth rate growth _ rate =2, the reproduction rate _ rate =1.8, the number of carnivorous plants nCPLant =10 and the number of preys nPrey =20 in a group, in an objective function SVM _ Acc, the maximum iteration times T =40, firstly, roughly optimizing parameters by a grid search method for parameters c and g of an optimization algorithm SVM, and determining the ranges of c and g to be 2 respectively -2 -2 8 I.e. 0.25-256 and 2 -5 -2 5 I.e. 0.03-32.
TABLE 1 test sample and corresponding target characteristic parameters
TABLE 2 comparison of classification methods
| Classification method | Test accuracy (%) |
| CPA-SVM | 93 |
| SVM | 82.5 |
| ELM | 85 |
It can be seen from table 1 that the average accuracy of the health state assessment classification obtained by using the two methods of the non-optimized SVM and the ELM is 82.5% and 85%, and it can be seen from fig. 5 and 6 that the classification accuracy obtained by using the CPA optimized SVM algorithm is improved more, and the effectiveness and accuracy of the method are verified.
It should be understood that although the specification has been described in terms of various embodiments, not every embodiment includes every single embodiment, and such description is for clarity purposes only, and it will be appreciated by those skilled in the art that the specification as a whole can be combined as appropriate to form additional embodiments as will be apparent to those skilled in the art. Equivalent embodiments or modifications that do not depart from the spirit of the technical spirit of the present invention should be included within the scope of the present invention.
Claims (8)
1. A two-dimensional piston pump health state assessment method based on variable rotation speed is characterized in that: which comprises the following steps:
s1, data acquisition: collecting vibration signals reflecting normal states and fault states of the two-dimensional piston pump as original state data;
s2, data preprocessing and analysis: carrying out GLCT time-frequency analysis on the original state data of the two-dimensional piston pump vibration signal acquired in the step S1, and extracting the instantaneous rotating speed frequency of the vibration signal;
s3, extracting angular domain vibration signals and order domain characteristic parameters, wherein the method comprises the following substeps:
s31, performing data fitting processing on the instantaneous rotating speed frequency extracted in the step S21, intercepting data in a frequency rising stage of the instantaneous frequency by adopting a segmented multi-order polynomial fitting method to perform data fitting to obtain a data fitting curve, and obtaining an order spectrum according to the fitting curve;
s32, for the angular domain vibration signal S (L) of the two-dimensional piston pump, if the length of the data sampled at equal angular intervals is L, the peak index Cf, the pulse index If and the kurtosis index K of the angular domain vibration signal S (L) are V The calculation formulas of (A) are respectively as follows:
wherein S is rms Is the root mean square value, S max In the form of a peak value, the peak value,
is the absolute average amplitude, β is the kurtosis;
root mean square value S rms The calculation formula of (2) is as follows:
peak value S max The calculation formula of (2) is as follows:
S max ＝max(|s(l)|)
absolute average amplitude
The calculation formula of (2) is as follows:
the kurtosis β is calculated as:
let the angular domain vibration signal s (L) order spectrum be Sl (m), where L =1,2, …, L, m is the order data length variable, and take m =1,2, …, D max Wherein D is max For maximum order, define the RMS value Y of the order spectrum 1 The following were used:
and use of Y 1 The vibration energy intensity of the whole order spectrum is measured;
s33, analyzing by using the amplitudes corresponding to the 11, 22, 33 and 44 orders of the order spectrum as order domain characteristic parameters:
wherein n is the number of the order spectrum data in the interval, Y 2 、Y 3 、Y 4 、Y 5 The amplitudes at orders 11, 22, 33 and 44 of the order spectrum, respectively;
s4, training and optimizing a fault diagnosis model, and specifically comprises the following substeps:
s41, constructing a fault diagnosis model: mixing Cf, if and K V 、Y 1 、Y 2 、Y 3 、Y 4 、Y 5 Constructing a fault diagnosis model as a training characteristic parameter of a CPA-SVM optimization classification algorithm, dividing a sample library into a training sample set and a test sample set according to a proportion, taking the training sample set as the input of the fault diagnosis model, and taking a state label as the output of the fault diagnosis model;
s42, accurately optimizing model parameters by using a CPA (cross correlation algorithm) optimization algorithm according to a punishment factor c and a radial cardinal number g in a large range under a target function of the classification accuracy of the SVM (support vector machine);
s43, performing fault classification model training by using the optimized CPA-SVM to obtain optimal model parameters best c and best g so as to obtain a trained and optimized fault diagnosis model;
s5, model verification: calling the trained and optimized fault diagnosis model, using the test sample set as the input of the trained and optimized fault diagnosis model, using the state label as the output of the trained and optimized fault diagnosis model, and verifying the comprehensive performance of the diagnosis model;
s6, model diagnosis: inputting the vibration signal of the two-dimensional piston pump into the trained and optimized fault diagnosis model, and outputting a state label value by the fault diagnosis model to obtain a health state classification and identification result.
2. The variable-speed-based two-dimensional piston pump state of health assessment method according to claim 1, characterized in that: the step S2 specifically includes the following substeps:
step S21, constructing an original STFT formula as follows:
wherein w (u-t') is a certain window; u is a certain short time;
step S22, on the basis of the STFT formula, a series of discrete demodulation operators are used for approximating the optimal demodulation operator
The STFT formula based on the discrete demodulation operator is obtained as follows:
step S23, for each TF point (t', w), if the demodulation operator is discrete
Close to the modulation component of the signal, TF around its IF indicates a higher energy concentration, whose amplitude | S (t ', w, c) | reaches a maximum value among all values, after which for each TF point, the optimal parameter c ' of the parameter c is obtained from the amplitude | S (t ', w, c) |:
the time-frequency representation of the proposed time-frequency analysis method is as follows:
GS(t',w)＝S(t',w,c')
the formula of the frequency spectrum calculation is as follows:
Spec(t',w)＝|GS(t',w)| 2
step S24, determining discrete demodulation operator
And (3) introducing a rotation parameter a to realize rotation on a time-frequency plane:
wherein, T s For sampling, F s Is the sampling frequency;
setting N values of the rotation parameter a, and dividing a time frequency plane into N +1 sections in an average way:
a＝-π/2+π/(N+1),-π/2+2·π/(N+1)...-π/2+N·π/(N+1)；
step S25, ensuring demodulation operator
All feasible modulation components in the signal can be described, and the STFT formula of the discrete demodulation operator is rewritten to obtain a GLCT formula:
and S26, carrying out GLCT time-frequency analysis according to the GLCT formula in the step S25, and extracting the instantaneous rotating speed frequency S (t', w, a) of the vibration signal.
3. The method for evaluating the health state of the variable-speed two-dimensional piston pump according to claim 1, wherein: step S31 specifically includes the following substeps:
s311, expressing the relation between the rotating speed and the first-order instantaneous frequency as follows:
wherein f (t) is the shaft rotation frequency, and P (t) is the reference shaft rotation speed;
s312, obtaining a fitting curve by adopting a piecewise multi-order polynomial fitting method:
R k (t)＝a k +b k t+c k t 2
wherein Rk (t) is a function of the speed curves of the individual sections,k is the sequence number of the segment, a k 、b k 、c k Is a polynomial coefficient;
s313, carrying out integral equation solving on the rotating speed fitting curve to obtain an equiangular interval interpolation phase discrimination time scale time sequence, wherein the solving process is as follows:
wherein, T n For key phase time scale, n is time scale sequence number, T 0 Is the initial fitting time; substituting the formula into a fitting curve formula, and obtaining a group of equiangular interval interpolation phase discrimination time scale time sequence T through effective algebraic solution n ：
S314, based on the equiangular interval interpolation phase discrimination time scale time sequence obtained in the step S313, smoothing is completed by using a Lagrange value method to obtain an angle domain signal;
and S315, carrying out Fourier transform on the angle domain signal data obtained in the step S314 to obtain an order spectrum.
4. The variable-speed-based two-dimensional piston pump state of health assessment method according to claim 1, characterized in that: 200 groups of data are acquired, and a training sample set and a test sample set are divided according to a proportion of 4:1, and the number of iterations in the group _ iter =20, the attraction rate absorption _ rate =0.8, the growth rate growth _ rate =2, the reproduction rate reproduction _ rate =1.8, the number of carnivorous plants ncpelant =10 and the number of prey =20 are defined.
5. The variable-speed-based two-dimensional piston pump state of health assessment method according to claim 1, characterized in that: step S42 specifically includes: 8 feature vectors obtained through time-frequency domain and order analysis are used as training feature parameters of a CPA-SVM optimization classification algorithm, target labels 1,2, 3 and 4 in the last column are four different wear states corresponding to the feature vectors according to the optimization classification algorithm, then the ranges of c and g are determined, and the c and g in the large range are accurately optimized by the CPA optimization algorithm under the objective function of the classification accuracy of the SVM.
6. The variable-speed-based two-dimensional piston pump state of health assessment method according to claim 1, characterized in that: in step S1, a vibration sensor mounted on the two-dimensional piston pump collects a vibration signal.
7. The variable-speed-based two-dimensional piston pump state of health assessment method according to claim 1, characterized in that: the optimal parameters are best c =224.6036, best g =31.5104.
8. The method for evaluating the health state of the variable-speed two-dimensional piston pump according to claim 1, wherein: the step S42 specifically includes:
the meat eating algorithm searches for a proper SVM parameter value in continuous iterative evolution, so that the accuracy of SVM classification is improved, the mathematical model can be simplified to search for an optimal decision vector [ c, g ], wherein c and g respectively represent a punishment parameter vector and a kernel function vector of the SVM, and the following aims are achieved by maximization:
max F＝SVM_Acc
in the formula, F represents an objective function, SVM _ Acc represents the classification accuracy of SVM, and M T Indicating the number of samples correctly classified into the corresponding category, M F Indicating the number of categories which are not correctly classified into corresponding categories; firstly, defining the iteration times group _ iter =20 in the group and the attraction rate of the attractionthe method comprises the steps of (1) performing a growth rate =0.8, a growth rate =2, a reproduction rate ratio =1.8, a carnivorous plant number nCPlant =10 and a prey number nPrey =20, performing rough optimization on parameters of an optimization algorithm SVM by a grid search method in an objective function SVM _ Acc with the maximum iteration number T =40, and determining the ranges of c and g to be 2 respectively -2 -2 8 I.e. 0.25-256 and 2 -5 -2 5 I.e. 0.03-32.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211400126.9A CN115795341A (en) | 2022-11-09 | 2022-11-09 | Two-dimensional piston pump health state assessment method based on variable rotating speed |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211400126.9A CN115795341A (en) | 2022-11-09 | 2022-11-09 | Two-dimensional piston pump health state assessment method based on variable rotating speed |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115795341A true CN115795341A (en) | 2023-03-14 |
Family
ID=85436370
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211400126.9A Pending CN115795341A (en) | 2022-11-09 | 2022-11-09 | Two-dimensional piston pump health state assessment method based on variable rotating speed |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115795341A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116738372A (en) * | 2023-08-15 | 2023-09-12 | 昆仑数智科技有限责任公司 | Rolling bearing fault diagnosis method, device and equipment for refining centrifugal pump |
-
2022
- 2022-11-09 CN CN202211400126.9A patent/CN115795341A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116738372A (en) * | 2023-08-15 | 2023-09-12 | 昆仑数智科技有限责任公司 | Rolling bearing fault diagnosis method, device and equipment for refining centrifugal pump |
| CN116738372B (en) * | 2023-08-15 | 2023-10-27 | 昆仑数智科技有限责任公司 | Rolling bearing fault diagnosis method, device and equipment for refining centrifugal pump |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Huo et al. | Incipient fault diagnosis of roller bearing using optimized wavelet transform based multi-speed vibration signatures | |
| Wang et al. | A novel weighted sparse representation classification strategy based on dictionary learning for rotating machinery | |
| CN109164343B (en) | Transformer fault diagnosis method based on characteristic information quantization and weighted KNN | |
| CN112036301B (en) | Driving motor fault diagnosis model construction method based on intra-class feature transfer learning and multi-source information fusion | |
| CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| CN103245907B (en) | A kind of analog-circuit fault diagnosis method | |
| CN103033362A (en) | Gear fault diagnosis method based on improving multivariable predictive models | |
| CN111122162B (en) | Industrial system fault detection method based on Euclidean distance multi-scale fuzzy sample entropy | |
| CN113340598B (en) | Rolling bearing intelligent fault diagnosis method based on regularized sparse model | |
| CN112507479B (en) | Oil drilling machine health state assessment method based on manifold learning and softmax | |
| CN110146279B (en) | Ship shafting unbalance fault diagnosis method based on vector evidence reasoning | |
| CN116380445B (en) | Equipment state diagnosis method and related device based on vibration waveform | |
| CN113268833A (en) | Migration fault diagnosis method based on deep joint distribution alignment | |
| CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
| CN115795341A (en) | Two-dimensional piston pump health state assessment method based on variable rotating speed | |
| CN114755017B (en) | Variable-speed bearing fault diagnosis method of cross-domain data driving unsupervised field shared network | |
| CN111898644A (en) | Intelligent identification method for health state of aerospace liquid engine under fault-free sample | |
| CN116950917A (en) | Large centrifugal compressor blade crack online diagnosis method based on blade tip timing | |
| Song et al. | Data and decision level fusion-based crack detection for compressor blade using acoustic and vibration signal | |
| CN115655717A (en) | Bearing fault diagnosis method based on depth domain generalization network | |
| CN113743010B (en) | Rolling bearing running state evaluation method based on EEMD energy entropy | |
| CN113869339A (en) | Deep learning classification model for fault diagnosis and fault diagnosis method | |
| CN112541524A (en) | BP-Adaboost multi-source information motor fault diagnosis method based on attention mechanism improvement | |
| CN111622815A (en) | Blade crack online measurement method based on blade tip timing and naive Bayes optimization | |
| CN115392109A (en) | LSTM multivariable time series anomaly detection method based on generative model |
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 |