CN113408076A - Small sample mechanical residual life prediction method based on support vector machine model - Google Patents

Small sample mechanical residual life prediction method based on support vector machine model Download PDF

Info

Publication number
CN113408076A
CN113408076A CN202110783436.2A CN202110783436A CN113408076A CN 113408076 A CN113408076 A CN 113408076A CN 202110783436 A CN202110783436 A CN 202110783436A CN 113408076 A CN113408076 A CN 113408076A
Authority
CN
China
Prior art keywords
support vector
residual life
vector machine
characteristic
machine
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
Application number
CN202110783436.2A
Other languages
Chinese (zh)
Inventor
黄贤振
王树凤
李禹雄
丁鹏飞
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.)
Yangzhou Super Machine Tool Co ltd
Original Assignee
Yangzhou Super Machine Tool Co ltd
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 Yangzhou Super Machine Tool Co ltd filed Critical Yangzhou Super Machine Tool Co ltd
Priority to CN202110783436.2A priority Critical patent/CN113408076A/en
Publication of CN113408076A publication Critical patent/CN113408076A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • 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
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/08—Probabilistic or stochastic CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Geometry (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention relates to a method for predicting the residual life of a small sample machine based on a support vector machine model, which comprises the following steps: selecting a plurality of measurement variables with obvious trends from the operation signal data of the mechanical product as characteristic quantities for describing the mechanical degradation process to obtain characteristic values of the characteristic quantities; preprocessing the characteristic value: smoothing the characteristic quantity of each mechanical degradation process by adopting a Butterworth filter algorithm; mapping the characteristic values into the same range by a normalization method; converting the mechanical residual life of each time point in the existing data into the proportion of the residual life to the total life, and normalizing the residual life to be in the range of [0, 1 ]; dividing the combination of a plurality of training sets and verification sets, wherein each combination corresponds to a support vector machine model; training each support vector machine model; and adopting each support vector machine model to predict the on-line residual life. The method fully utilizes limited mechanical operation data and can realize accurate prediction of the residual service life of the machine.

Description

Small sample mechanical residual life prediction method based on support vector machine model
Technical Field
The invention relates to a method for predicting the residual life of a small sample machine based on a support vector machine model, and belongs to the field of residual life prediction of mechanical products.
Background
The remaining life of a machine in the prior art is defined as the length of time the machine has elapsed from the current point in time until it completely fails. Residual life prediction is an important link in fault diagnosis and Health Management (PHM). Accurate residual life prediction can provide effective suggestions and schemes for maintenance and replacement of machinery, thereby greatly improving the reliability and stability of the system.
The method for predicting the residual life of the machine based on data driving is the mainstream of the current research. The data driving method adopts a machine learning algorithm to establish a mapping relation between the monitoring data and the residual life of the machine. Common methods include Artificial Neural Networks (ANN), neuro-fuzzy systems (NF systems), and Deep Learning (DL), among others. However, the current mechanical learning algorithm needs a large number of high-quality sample points to train the model, and when the number of samples is small or the overall quality of the samples is poor, the error of the final prediction result is large. In engineering practice, it often takes days or even months for a machine to go from normal operation to complete failure, and therefore obtaining a full-life machine operation signal data collection is often costly. In addition, environmental noise and machine manufacturing errors can cause a large amount of interference to the observed signals, reducing the quality of the data. A Support Vector Machine (SVM) is a machine learning algorithm currently applicable to small sample conditions, and has a certain application in the remaining life of a machine.
The disadvantages are that: due to improper selection of model parameters and large difference of signals of different elements, the prediction method based on the SVM can generate over-fitting and under-fitting problems, and the prediction accuracy is reduced.
Disclosure of Invention
The invention aims to overcome the problems in the prior art and provide a method for predicting the residual life of a small sample machine based on a support vector machine model, which fully utilizes limited machine operation data, establishes a plurality of SVM models, calculates optimal model parameters through an optimization algorithm, introduces a weight value to automatically update the algorithm to distribute the model proportion, and thus realizes the accurate prediction of the residual life of the machine.
In order to solve the technical problems, the invention provides a method for predicting the residual life of a small sample machine based on a support vector machine model, which comprises the following steps:
step 1, analyzing operation signal data of a mechanical product, selecting a plurality of measurement variables with obvious trends as characteristic quantities for describing a mechanical degradation process, and obtaining characteristic values of the characteristic quantities;
step 2, preprocessing the selected characteristic values: smoothing the characteristic quantity of each mechanical degradation process by adopting a Butterworth filter algorithm to reduce random noise; the characteristic values of the characteristic quantities are mapped into the same range through a normalization method, so that the support vector machine model is conveniently trained;
step 3, converting the mechanical residual life of each time point in the existing data into the proportion of the residual life to the total life, and normalizing the residual life to be in the range of [0, 1 ];
step 4, dividing a plurality of combinations of training sets and verification sets according to the scale and the characteristics of the existing data, wherein each combination corresponds to a support vector machine model;
step 5, training each support vector machine model;
and 6, adopting each support vector machine model to predict the on-line residual life.
As a preferred embodiment of the present invention, the specific process of smoothing the feature quantity of each mechanical degradation process by using the butterworth filter algorithm in step 2 is as follows: filtering the original characteristic quantity, wherein the formula is as follows:
Figure BDA0003158119820000021
in the formula: n represents the filtering order; omegacRepresents the cut-off frequency; g0Representing the benefit at zero frequency, components with frequencies above the cutoff frequency will be filtered, while components below the cutoff frequency will be retained.
As a further preferable aspect of the present invention, the formula according to which the normalization processing in step 3 is based is as follows:
y＝(ymax-ymin)(x-xmin)/(xmax-xmin)+yminin the formula: y is the result of the normalization; x is the original data value; y ismaxAnd yminRespectively the upper and lower boundaries of the range to be normalized; x is the number ofmaxAnd xminThe upper and lower bounds of the original data value.
As a further preferable embodiment of the present invention, step 5 specifically includes the following steps:
step 5-1, taking the preprocessed characteristic value as an input quantity theta of the model, and taking the proportion of the residual life to the total life as an output quantity R of the model;
step 5-2, setting a multi-index optimization target to enable the optimization direction to cover the prediction precision and the fitting speed of the support vector machine;
5-3, performing parameter optimization on each individual support vector machine by adopting an artificial bee colony optimization algorithm to obtain model parameters with optimal regression and prediction effects;
and 5-4, verifying the validity of each model by testing the data of the set.
As a further preferred embodiment of the present invention, the input amount θ ═ θ in step 5-11，θ2，...，θN]Wherein N is the number of the characteristic quantities,
Figure BDA0003158119820000022
the j characteristic quantity containing K data; output quantity R ═ R1，r2，L，rK]a/T, wherein riIs a point of time tiThe remaining life of the machine, T, is the total life of the machine.
As a further preferable scheme of the present invention, in step 5-2, the multi-index optimization target is set, and the target parameters are as follows:
Figure BDA0003158119820000031
in the formula, mse is the mean square error of the regression effect, and it is assumed that the actual value and the regression result in the training set are λ and
Figure BDA0003158119820000032
then the equation for mse is:
Figure BDA0003158119820000033
d represents an error index describing the prediction effect of the verification set, and v is a corresponding weight.
As a further preferable scheme of the invention, the step 5-3 specifically comprises the following steps:
step 5-31: setting an initial population scale D, setting a maximum iteration number Max, and randomly generating an initial solution vector, wherein the formula is as follows: omegadL + rand (0,1) × (u-l), where ω isdD is more than or equal to 1 and less than or equal to D, the generated solution vector comprises H elements, and u and 1 are the upper boundary and the lower boundary of the solution vector respectively;
step 5-32: searching for a new solution vector around the initial solution vector according to the following formula:
ω′d，m＝ωd，m+φ(ωd，m-ωb，m) Wherein, ω isd，mIs the solution vector omegadThe m-th element of (1), ωb，mIs the solution vector omegabThe m-th element of (1); m is more than or equal to 1 and less than or equal to H; phi is [0, 1]A random number within a range;
step 5-33: and calculating the fitness fit of the new solution vector, wherein the formula is as follows:
Figure BDA0003158119820000034
where fit (ω)d) Is an optimized objective function;
step 5-34: giving each solution vector an acceptance probability according to the fitness, and selecting an ideal solution according to the probability, wherein the formula is as follows:
Figure BDA0003158119820000035
step 5-35: and (5) repeating the steps 5-31 to 5-34, and updating the optimal solution in each iteration until the maximum iteration number is reached or the optimal value is not changed in a plurality of iterations, namely obtaining the model parameters with optimal regression and prediction effects.
As a further preferable embodiment of the present invention, step 6 specifically includes the following steps:
6-1, performing sliding average processing on the mechanical signal in the current operation in real time to reduce random noise;
6-2, determining the average extreme value of each characteristic quantity according to the existing data set, and carrying out normalization processing on the mechanical signal in the current operation;
6-3, calculating Euclidean distance between each characteristic quantity and the corresponding characteristic quantity in the training sample to evaluate the similarity between the current machine and the machines in the training set;
step 6-4, according to the similarity between the training set machine and the current observation machine in each support vector machine model, giving corresponding weight to each single model, and obtaining the current residual life proportion through weighted average to obtain a residual life prediction result;
and 6-5, carrying out averaging processing on the residual life prediction result in a time period, and further reducing the prediction error.
As a further preferable aspect of the present invention, the calculation formula of the euclidean distance in step 6-3 is as follows:
Figure BDA0003158119820000041
wherein, θ' and θ are the characteristic value of the current measuring machine and the characteristic value of a certain machine in the training set, respectively.
As a further preferable embodiment of the present invention, the calculation formula of the remaining life prediction result in step 6-4 is as follows:
Figure BDA0003158119820000042
wherein, wiIs the weight of the ith support vector machine, riIs the prediction result of the ith support vector machine.
Compared with the prior art, the invention has the following beneficial effects: the small sample mechanical residual life prediction method based on the support vector machine can more effectively utilize limited data samples to establish the support vector machine model, is suitable for engineering application with difficulty in obtaining the data samples, and has wider applicability compared with other machine learning algorithms needing large scale and high quality. And a multi-index construction optimization target is introduced, the optimal model parameters are obtained through a manual bee colony optimization algorithm, the prediction precision of each support vector machine model is improved, and the universality of the model is enhanced. Meanwhile, a weight calculation method based on Euclidean distance is introduced, and the weight corresponding to each support vector machine is obtained according to the similarity of the characteristic quantity and the training sample, so that the phenomena of over-fitting and under-fitting are reduced, the prediction efficiency and stability are further improved, and the method has important engineering value.
Drawings
The invention will be described in further detail with reference to the following drawings and detailed description, which are provided for reference and illustration purposes only and are not intended to limit the invention.
FIG. 1 is a flow chart of an embodiment of the present invention.
FIG. 2 is a flow chart of the artificial bee colony algorithm of the present invention.
FIG. 3 is an online prediction effect of a single support vector machine on the remaining life of a machine.
FIG. 4 is an online prediction effect of a weighted support vector machine on the remaining life of a machine.
Detailed Description
As shown in fig. 1, the method for predicting the remaining life of a small sample machine based on a support vector machine model of the invention comprises the following steps:
step 1, analyzing operation signal data of a mechanical product, selecting a plurality of measurement variables with obvious trends as characteristic quantities for describing a mechanical degradation process, and obtaining characteristic values of the characteristic quantities;
step 2, preprocessing the selected characteristic values: smoothing the characteristic quantity of each mechanical degradation process by adopting a Butterworth filter algorithm to reduce random noise; the characteristic values of the characteristic quantities are mapped into the same range through a normalization method, so that the support vector machine model is conveniently trained;
step 3, converting the mechanical residual life of each time point in the existing data into the proportion of the residual life to the total life, and normalizing the residual life to be in the range of [0, 1 ];
step 4, dividing a plurality of combinations of training sets and verification sets according to the scale and the characteristics of the existing data, wherein each combination corresponds to a support vector machine model;
step 5, training each support vector machine model;
and 6, adopting each support vector machine model to predict the on-line residual life.
The specific process of smoothing the characteristic quantity of each mechanical degradation process by adopting the Butterworth filter algorithm in the step 2 is as follows: filtering the original characteristic quantity, wherein the formula is as follows:
Figure BDA0003158119820000051
in the formula: n represents the filtering order; omegacRepresents the cut-off frequency; g0Representing the benefit at zero frequency, components with frequencies above the cutoff frequency will be filtered, while components below the cutoff frequency will be retained.
The formula according to which the normalization process in step 3 is based is as follows:
y＝(ymax-ymin)(x-xmin)/(xmax-xmin)+ymin (2)
in the formula: y is the result of the normalization; x is the original data value; y ismaxAnd yminRespectively the upper and lower boundaries of the range to be normalized; x is the number ofmaxAnd xminThe upper and lower bounds of the original data value.
The step 5 specifically comprises the following steps:
step 5-1, using the preprocessed characteristic value as the input quantity theta of the model, wherein theta is [ theta ]1，θ2，...，θN]Wherein N is the number of the characteristic quantities,
Figure BDA0003158119820000052
the j characteristic quantity containing K data;
the ratio of the remaining life to the total life is also used as the output R of the model, R ═ R1，r2，L，rK]a/T, wherein riIs a point of time tiThe remaining life of the machine, T, is the total life of the machine.
Step 5-2, setting a multi-index optimization target to enable the optimization direction to cover the prediction precision and the fitting speed of the support vector machine; the target parameters were as follows:
Figure BDA0003158119820000053
in the formula, mse is the mean square error of the regression effect, and it is assumed that the actual value and the regression result in the training set are λ and
Figure BDA0003158119820000054
then the equation for mse is:
Figure BDA0003158119820000061
in formula (3), d represents an error index describing the prediction effect of the verification set, and v represents a corresponding weight and represents each indexAnd marking the influence of the optimization result. In the method, three error indexes d are selected to evaluate the prediction effect of the verification set. Suppose that the actual values and the predicted results in the verification set are gamma and
Figure BDA0003158119820000062
the three index calculation formulas are respectively as follows:
d1 Root Mean Square Error (RMSE):
Figure BDA0003158119820000063
d2mean Relative Error (MRE):
Figure BDA0003158119820000064
d3degree of fit (Convergence):
Figure BDA0003158119820000065
wherein the content of the first and second substances,
Figure BDA0003158119820000066
Figure BDA0003158119820000067
in the formulae (7) to (9), tFPTAt an initial time point, tiRepresents the ith time point; m (t)i) Is tiRelative error of the true value and the predicted value at the time point.
And 5-3, performing parameter optimization on each individual support vector machine by adopting an artificial bee colony optimization algorithm to obtain model parameters with optimal regression and prediction effects, wherein the method specifically comprises the following steps:
step 5-31: as shown in fig. 2, an initial population size D is set, a maximum number of iterations Max is set, and an initial solution vector is randomly generated, where the formula is as follows:
ωd＝l+rand(0,1)×(u-l) (10)
wherein, ω isdD is more than or equal to 1 and less than or equal to D, H elements are contained, and u and l are the upper and lower bounds of the solution vector respectively.
Step 5-32: searching for a new solution vector around the initial solution vector according to equation (11):
ω'd,m＝ωd,m+φ(ωd,m-ωb,m) (11)
wherein, ω isd,mIs the solution vector omegadM (m is more than or equal to 1 and less than or equal to H) element, omegab,mIs the solution vector omegabThe m (m is more than or equal to 1 and less than or equal to H) element; phi is [0, 1]A random number within a range;
step 5-33: and calculating the fitness fit of the new solution vector, wherein the formula is as follows:
Figure BDA0003158119820000071
wherein f (ω)d) Is an optimized objective function, namely formula (3);
step 5-34: giving each solution vector an acceptance probability according to the fitness, and selecting an ideal solution according to the probability, wherein the formula is as follows:
Figure BDA0003158119820000072
step 5-35: and (5) repeating the steps 5-31 to 5-34, and updating the optimal solution in each iteration until the maximum iteration number is reached or the optimal value is not changed in a plurality of iterations, namely obtaining the model parameters with optimal regression and prediction effects.
And 5-4, verifying the validity of each model by testing the data of the set.
The step 6 specifically comprises the following steps:
6-1, performing sliding average processing on the mechanical signal in the current operation in real time to reduce random noise;
6-2, determining the average extreme value of each characteristic quantity according to the existing data set, and carrying out normalization processing on the mechanical signal in the current operation;
6-3, calculating Euclidean distance between each characteristic quantity and the corresponding characteristic quantity in the training sample to evaluate the similarity between the current machine and the machines in the training set; the calculation formula of the euclidean distance is as follows:
Figure BDA0003158119820000073
wherein, θ' and θ are respectively the characteristic value of the current measuring machine and the characteristic value of a certain machine in the training set, and the characteristic values can reflect the similarity between the variation.
6-4, according to the similarity between the training set machine and the current observation machine in each support vector machine model, giving a corresponding weight to each single model, and obtaining a current residual life ratio through weighted average to obtain a residual life prediction result, wherein the calculation formula is as follows:
Figure BDA0003158119820000081
wherein, wiIs the weight of the ith support vector machine, riIs the prediction result of the ith support vector machine.
And 6-5, carrying out averaging processing on the residual life prediction result in a time period, and further reducing the prediction error.
Next, a remaining life prediction is performed for the CMAPSS turbine engine data set, comprising the steps of:
step 1, selecting a proper characteristic value:
the CMAPSS engine data contains 14 indicators describing the state of health of the engine, including: fan inlet temperature, compressor outlet temperature, turbine outlet temperature, fan inlet pressure, bypass duct pressure, relative blade speed, core speed, engine pressure ratio, and the like. The pressure ratio, the static pressure at the outlet of the compressor and the change trend of the relative blade rotating speed are more obvious, and the degradation process of the engine can be better reflected, so that the three variables are selected as characteristic values by the method.
And 2, 3, setting a proper filtering order and a proper cut-off frequency according to the frequency characteristics of different characteristic values, filtering the characteristic values according to a formula (1), and normalizing the characteristic values according to a formula (2).
Step 4, dividing a training set and a verification set: to simulate small sample conditions, only 5 of the 100 samples in the training set were selected as known data, and the other samples were selected as unknown data. A total of 10 support vector machine models can be trained by selecting 3 groups from 5 groups of samples as a training set and the remaining 2 groups as a verification set.
Step 5, training each support vector machine model:
step 5-1, determining input quantity and output quantity: and constructing a 3 xK input quantity matrix according to the preprocessed characteristic values, wherein K is the total number of time points of 3 groups of samples. Because the engine in the CMAPSS data set is in a steady state operation state for an initial period of time, values of the residual life greater than 125 are unified into 125 when the output quantity is constructed, and then the residual life proportion is calculated correspondingly.
Step 5-2, setting a multi-index optimization target according to the required optimization effect: since in engineering practice, overestimation of the remaining lifetime will have more serious consequences than underestimation, it is desirable to avoid overestimation as much as possible. When the objective function is constructed, a higher weight can be distributed to the index of the fitting degree, so that the prediction result is converged as soon as possible.
And 5-3, optimizing the model parameters by adopting an artificial bee colony algorithm. For the CMAPSS engine model, the radial basis kernel function is selected when the support vector machine is trained, so the parameters to be determined in the model include a penalty coefficient C and a radial basis kernel function width g. The population size is set to be 20, the maximum iteration number is set to be 50, and optimization is stopped when the optimal value does not change in 10 iterations. Optimizing the parameters according to the formula (10-13), and finally obtaining the optimal parameters of 10 support vector machine models, wherein the result is shown in table 1;
TABLE 1 support vector machine model optimization parameters
Figure BDA0003158119820000082
Figure BDA0003158119820000091
And 5-4, verifying the validity of the model according to the test set: the regression effect and the test effect on the validation set for each model are shown in table 2.
TABLE 2 support vector machine model Effect
Figure BDA0003158119820000092
And 6, adopting each support vector machine model to predict the residual life:
and 6-1 and 2, performing sliding average on the characteristic values obtained by measurement, and reducing random noise of the characteristic values. According to the data in the training set, the average maximum values of the three feature vectors are respectively: 2388.2587, 8.5309, and 554.3199, normalizing the three feature quantities according to the three average maximum values, respectively, to construct input quantities.
And 6-3, calculating Euclidean distance evaluation of the characteristic values and the characteristic values in the training set to evaluate similarity, wherein tables 3 and 4 show Euclidean distances of the three characteristic quantities and the corresponding characteristic values of each engine individual in the training set at different time points.
TABLE 3 Euclidean distance of characteristic values under 20-cycle operation
Figure BDA0003158119820000093
TABLE 4 Euclidean distance of eigenvalues under 150 cycle loop of operation
Figure BDA0003158119820000101
And 6-4, determining the weight of each support vector machine model, taking the time point 20 cycle period: 150 cycle period as an example, the reciprocal of each characteristic value Euclidean distance can reflect the similarity, so that the similarity between the current test engine and the engines 1-5 can be written as follows:
20 cycle period:
Figure BDA0003158119820000102
Figure BDA0003158119820000103
Figure BDA0003158119820000104
Figure BDA0003158119820000105
Figure BDA0003158119820000106
150 cycle period:
Figure BDA0003158119820000107
Figure BDA0003158119820000108
Figure BDA0003158119820000109
Figure BDA00031581198200001010
Figure BDA00031581198200001011
each support vector machine contains data of 3 engines as training samples, so the applicability of each support vector machine to the current test engine can be determined by adding the similarity s, and the weight of each support vector machine can be determined by the proportion of the applicability. Thus, at 20 cycle period, the fitness and weight of the model are as shown in table 5:
TABLE 5 suitability and weight of each model for 20 cycle run
Figure BDA00031581198200001012
Figure BDA0003158119820000111
At 150 cycle period, the fitness and weight of the model are shown in table 6:
TABLE 6 suitability and weight for each model over a 150-cycle run
Figure BDA0003158119820000112
The online prediction effect of a single support vector machine on the residual life of the engine is shown in FIG. 3, and the online prediction effect of the weighted support vector machine on the residual life of the engine is shown in FIG. 4. Wherein:
the predicted remaining life ratios of the 10 support vector machine models when running for 20 cycles are respectively: 0.9870,0.9431,0.9852,0.9193,0.9998,1.0009,0.9263,0.9535,0.9060,1.0533. According to equation (13), the weighted remaining life ratio results as: 0.9696.
the predicted residual life ratios of the 10 support vector machines when operating for the 150-cycle are respectively as follows: 0.7986,0.6720,0.8327,0.5157,0.6356,0.5680,0.6677,08355,0.6606, 05564. According to equation (13), the weighted remaining life ratio results as: 0.6794.
and 6-5, averaging the prediction result in order to further reduce the prediction error. Firstly, converting the proportion of the residual life into an actual life value, and averaging the residual life values of the five latest predicted time points to obtain a final predicted value of the current time point. The remaining life at 20 cycle and 150 cycle can be calculated as follows:
Figure BDA0003158119820000113
Figure BDA0003158119820000114
the actual remaining life at the two time points was 125 and 102, and it can be seen that the predicted results are very close to the actual values.
Table 7 shows the error evaluation of the predicted remaining life values at all time points.
TABLE 7 residual Life prediction error
Figure BDA0003158119820000121
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention. In addition to the above embodiments, the present invention may have other embodiments. All technical solutions formed by adopting equivalent substitutions or equivalent transformations fall within the protection scope of the claims of the present invention. Technical features of the present invention which are not described may be implemented by or using the prior art, and will not be described herein.

Claims (10)

1. A small sample mechanical residual life prediction method based on a support vector machine model is characterized by comprising the following steps:
step 1, analyzing operation signal data of a mechanical product, selecting a plurality of measurement variables with obvious trends as characteristic quantities for describing a mechanical degradation process, and obtaining characteristic values of the characteristic quantities;
step 2, preprocessing the selected characteristic values: smoothing the characteristic quantity of each mechanical degradation process by adopting a Butterworth filter algorithm to reduce random noise; the characteristic values of the characteristic quantities are mapped into the same range through a normalization method, so that the support vector machine model is conveniently trained;
step 3, converting the mechanical residual life of each time point in the existing data into the proportion of the residual life to the total life, and normalizing the residual life to be in the range of [0, 1 ];
step 4, dividing a plurality of combinations of training sets and verification sets according to the scale and the characteristics of the existing data, wherein each combination corresponds to a support vector machine model;
step 5, training each support vector machine model;
and 6, adopting each support vector machine model to predict the on-line residual life.
2. The method for predicting the residual life of the small sample machine based on the support vector machine model according to claim 1, wherein the smoothing of the feature quantity of each mechanical degradation process by using the butterworth filter algorithm in the step 2 is specifically performed by: filtering the original characteristic quantity, wherein the formula is as follows:
Figure FDA0003158119810000011
in the formula: n represents the filtering order; omegacRepresents the cut-off frequency; g0Representing benefit at zero frequency, higher than cut-off frequencyThe components of the rate will be filtered while the components below the cut-off frequency will be retained.
3. The method for predicting the residual life of the small sample machine based on the support vector machine model according to claim 1, wherein the normalization in the step 3 is based on the following formula:
y＝(ymax-ymin)(x-xmin)/(xmax-xmin)+ymin，
in the formula: y is the result of the normalization; x is the original data value; y ismaxAnd yminRespectively the upper and lower boundaries of the range to be normalized; x is the number ofmaxAnd xminThe upper and lower bounds of the original data value.
4. The method for predicting the residual life of the small sample machine based on the support vector machine model according to claim 1, wherein the step 5 specifically comprises the following steps:
step 5-1, taking the preprocessed characteristic value as an input quantity theta of the model, and taking the proportion of the residual life to the total life as an output quantity R of the model;
step 5-2, setting a multi-index optimization target to enable the optimization direction to cover the prediction precision and the fitting speed of the support vector machine;
5-3, performing parameter optimization on each individual support vector machine by adopting an artificial bee colony optimization algorithm to obtain model parameters with optimal regression and prediction effects;
and 5-4, verifying the validity of each model by testing the data of the set.
5. The method for predicting the residual life of the small sample machine based on the support vector machine model according to claim 4, wherein: the input amount θ [ θ ] described in step 5-11，θ2，...，θN]Wherein N is the number of the characteristic quantities,
Figure FDA0003158119810000021
for containing K dataThe jth feature quantity; output quantity R ═ R1,r2,L,rK]a/T, wherein riIs a point of time tiThe remaining life of the machine, T, is the total life of the machine.
6. The method for predicting the residual life of the small sample machine based on the support vector machine model according to claim 4, wherein the multi-index optimization target is set in step 5-2, and the target parameters are as follows:
Figure FDA0003158119810000022
in the formula, mse is the mean square error of the regression effect, and it is assumed that the actual value and the regression result in the training set are λ and
Figure FDA0003158119810000023
then the equation for mse is:
Figure FDA0003158119810000024
d represents an error index describing the prediction effect of the verification set, and v is a corresponding weight.
7. The method for predicting the residual life of the small sample machine based on the support vector machine model according to claim 4, wherein the step 5-3 specifically comprises the following steps:
step 5-31: setting an initial population scale D, setting a maximum iteration number Max, and randomly generating an initial solution vector, wherein the formula is as follows: omegadL + rand (0,1) × (u-l), where ω isdD is more than or equal to 1 and less than or equal to D, the generated solution vector comprises H elements, and u and l are the upper boundary and the lower boundary of the solution vector respectively;
step 5-32: searching for a new solution vector around the initial solution vector according to the following formula:
ω'd,m＝ωd,m+φ(ωd,m-ωb,m) Wherein, ω isd,mIs the solution vector omegadThe m-th element of (1), ωb,mIs the solution vector omegabM is more than or equal to 1 and less than or equal to H; phi is [0, 1]Random number within range；
Step 5-33: and calculating the fitness fit of the new solution vector, wherein the formula is as follows:
Figure FDA0003158119810000025
where fit (ω)d) Is an optimized objective function;
step 5-34: giving each solution vector an acceptance probability according to the fitness, and selecting an ideal solution according to the probability, wherein the formula is as follows:
Figure FDA0003158119810000031
step 5-35: and (5) repeating the steps 5-31 to 5-34, and updating the optimal solution in each iteration until the maximum iteration number is reached or the optimal value is not changed in a plurality of iterations, namely obtaining the model parameters with optimal regression and prediction effects.
8. The method for predicting the residual life of the small sample machine based on the support vector machine model according to claim 1, wherein the step 6 specifically comprises the following steps:
6-1, performing sliding average processing on the mechanical signal in the current operation in real time to reduce random noise;
6-2, determining the average extreme value of each characteristic quantity according to the existing data set, and carrying out normalization processing on the mechanical signal in the current operation;
6-3, calculating Euclidean distance between each characteristic quantity and the corresponding characteristic quantity in the training sample to evaluate the similarity between the current machine and the machines in the training set;
step 6-4, according to the similarity between the training set machine and the current observation machine in each support vector machine model, giving corresponding weight to each single model, and obtaining the current residual life proportion through weighted average to obtain a residual life prediction result;
and 6-5, carrying out averaging processing on the residual life prediction result in a time period, and further reducing the prediction error.
9. The method for predicting the residual life of the small sample machine based on the support vector machine model according to claim 8, wherein: the calculation formula of the Euclidean distance in the step 6-3 is as follows:
Figure FDA0003158119810000032
wherein, θ' and θ are the characteristic value of the current measuring machine and the characteristic value of a certain machine in the training set, respectively.
10. The method for predicting the residual life of the small sample machine based on the support vector machine model according to claim 8, wherein: the calculation formula of the residual life prediction result in the step 6-4 is as follows:
Figure FDA0003158119810000033
wherein, wiIs the weight of the ith support vector machine, riIs the prediction result of the ith support vector machine.
CN202110783436.2A 2021-07-12 2021-07-12 Small sample mechanical residual life prediction method based on support vector machine model Pending CN113408076A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110783436.2A CN113408076A (en) 2021-07-12 2021-07-12 Small sample mechanical residual life prediction method based on support vector machine model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110783436.2A CN113408076A (en) 2021-07-12 2021-07-12 Small sample mechanical residual life prediction method based on support vector machine model

Publications (1)

Publication Number Publication Date
CN113408076A true CN113408076A (en) 2021-09-17

Family

ID=77686123

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110783436.2A Pending CN113408076A (en) 2021-07-12 2021-07-12 Small sample mechanical residual life prediction method based on support vector machine model

Country Status (1)

Country Link
CN (1) CN113408076A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114996245A (en) * 2022-04-07 2022-09-02 济南大学 Data compression method applied to cement production big data

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114996245A (en) * 2022-04-07 2022-09-02 济南大学 Data compression method applied to cement production big data

Similar Documents

Publication Publication Date Title
US11436395B2 (en) Method for prediction of key performance parameter of an aero-engine transition state acceleration process based on space reconstruction
CN111414977B (en) Weighted integration temperature sensitive point combination selection method for machine tool spindle thermal error modeling
CN111369070B (en) Multimode fusion photovoltaic power prediction method based on envelope clustering
CN106875037B (en) Wind power prediction method and device
CN109063276B (en) Wind power plant dynamic equivalent modeling method suitable for long-time domain wind speed fluctuation
CN108804850B (en) Method for predicting parameters of aircraft engine in transient acceleration process based on spatial reconstruction
CN116150897A (en) Machine tool spindle performance evaluation method and system based on digital twin
CN112818604A (en) Wind turbine generator risk degree assessment method based on wind power prediction
CN114583767B (en) Data-driven wind power plant frequency modulation response characteristic modeling method and system
CN110007660B (en) Online soft measurement method for transient equivalent thermal stress of steam turbine set of thermal power plant
CN113408076A (en) Small sample mechanical residual life prediction method based on support vector machine model
CN113343562B (en) Fan power prediction method and system based on hybrid modeling strategy
CN108428012B (en) Fan noise prediction method for optimizing neural network
CN114329347A (en) Method and device for predicting metering error of electric energy meter and storage medium
CN109614722B (en) Modeling method for scroll engine full-state parameters based on fuzzy logic
CN117232809A (en) Fan main shaft fault pre-diagnosis method based on DEMATEL-ANP-CRITIC combined weighting
CN116777109A (en) Wind generating set equipment state evaluation method and system
CN115907192A (en) Method and device for generating wind power fluctuation interval prediction model and electronic equipment
CN113048012B (en) Wind turbine generator yaw angle identification method and device based on Gaussian mixture model
CN111950131B (en) Wind power plant output equivalent aggregation model construction method considering electricity limiting factors
Duan et al. Study on performance evaluation and prediction of mixed-flow hydraulic turbine units under variable operating conditions
CN116738759B (en) Method and device for designing and operating equipment, computer equipment and readable storage medium
CN113722970B (en) Photovoltaic power ultra-short-term online prediction method
CN117893030B (en) Power system risk early warning method based on big data
CN113465930B (en) Gas turbine multi-sensor fault detection method based on hybrid method

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