CN111307453B - Transmission system fault diagnosis method based on multi-information fusion - Google Patents

Transmission system fault diagnosis method based on multi-information fusion Download PDF

Info

Publication number
CN111307453B
CN111307453B CN202010199373.1A CN202010199373A CN111307453B CN 111307453 B CN111307453 B CN 111307453B CN 202010199373 A CN202010199373 A CN 202010199373A CN 111307453 B CN111307453 B CN 111307453B
Authority
CN
China
Prior art keywords
feature
transmission system
signals
extracting
vector
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.)
Active
Application number
CN202010199373.1A
Other languages
Chinese (zh)
Other versions
CN111307453A (en
Inventor
张品佳
吴志良
袁巍
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.)
Lonston Technology Beijing Co ltd
Original Assignee
Lonston Technology Beijing 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 Lonston Technology Beijing Co ltd filed Critical Lonston Technology Beijing Co ltd
Priority to CN202010199373.1A priority Critical patent/CN111307453B/en
Publication of CN111307453A publication Critical patent/CN111307453A/en
Application granted granted Critical
Publication of CN111307453B publication Critical patent/CN111307453B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/02—Gearings; Transmission mechanisms
    • G01M13/028—Acoustic or vibration analysis

Abstract

The invention relates to a transmission system fault diagnosis method based on multi-information fusion, which comprises the steps of obtaining multiple groups of original signals of a transmission system in normal and fault states; preprocessing each acquired signal; respectively extracting first feature vectors, and performing normalization processing and feature fusion to obtain training samples; constructing a support vector machine model, and training by using a training sample; and (3) acquiring each signal of the transmission system in real time for preprocessing, extracting a second feature vector and normalizing, diagnosing by adopting the trained support vector machine diagnosis model after feature fusion, and outputting a fault detection result. The invention collects multiple signals to extract the characteristics and perform characteristic fusion, can effectively overcome the problem of low accuracy rate caused by diagnosis by single characteristic, obtains a diagnosis result which is more consistent with the real running state of the motor, and effectively improves the fault diagnosis rate of the transmission system.

Description

Transmission system fault diagnosis method based on multi-information fusion
Technical Field
The invention relates to the technical field of fault diagnosis, in particular to a transmission system fault diagnosis method based on multi-information fusion.
Background
The fault diagnosis technology of the transmission system is based on mathematical modeling, collects data through data acquisition and data mining, extracts fault characteristics by using a signal processing technology, and performs fault diagnosis through the characteristics and development trend of the data.
The existing technical scheme mainly comprises an expert system, a fuzzy theory and the like; the expert system is a program system with a great deal of special knowledge and experience, firstly needs to collect a great deal of knowledge and data in advance to form an independent knowledge base system, then carries out reasoning and judgment according to the logical relationship established by the knowledge base system, and finally programs the logical relationship to simulate the decision process of an expert to realize the fault diagnosis process.
The existing fault diagnosis method only combines a certain fault feature set or a certain pattern recognition algorithm, and only can find the optimal fault classification method from a single angle, so that the one-sidedness is bound to influence the reliability of fault diagnosis of the transmission system, and the accuracy of fault diagnosis is low.
In fault diagnosis, as transmission structures such as fans, pumps, compressors, gears and motors are increasingly large and complicated, and single or single-domain characteristics have certain contingencies, so that all fault characteristics of a transmission system cannot be included, detection omission or misjudgment is often caused in fault diagnosis. The advantages of each characteristic domain and each diagnosis method can be reserved by performing multi-domain characteristic extraction and multi-decision result fusion on the equipment, the purpose of improving the fault diagnosis precision is achieved, the transmission system is comprehensively analyzed and subjected to fault diagnosis by adopting an information fusion method, and the influence of widely-existing uncertain factors on diagnosis can be reduced.
Disclosure of Invention
In order to improve the precision and reliability of fault diagnosis of the transmission system, the invention provides a fault diagnosis method of the transmission system based on multi-information fusion.
In order to achieve the above object, the present invention provides a transmission system fault diagnosis method based on multi-information fusion, which includes:
acquiring a plurality of groups of original signals of the transmission system in normal and fault states, wherein the original signals comprise current signals, vibration signals and 2-4 signals in voltage signals, rotating speed signals, temperature signals and noise signals; preprocessing each acquired original signal; respectively extracting first feature vectors; normalizing each first feature vector; performing feature fusion on the normalized first feature vector by adopting a self-adaptive weighting algorithm, and taking the fused first feature as a training sample;
constructing a support vector machine model, and training the support vector machine model by using a training sample to output a diagnosis result;
acquiring a current signal, a vibration signal and the 2 to 3 signals of the transmission system in real time; preprocessing each signal acquired in real time, respectively extracting and normalizing second feature vectors, and performing feature fusion on each normalized second feature vector; and diagnosing the fused feature vector by adopting the trained support vector machine diagnosis model, and outputting a fault detection result.
Further, the preprocessing of the acquired signals and the signals acquired in real time comprises filtering and abnormal value processing.
Further, extracting the first feature vector comprises extracting the feature vector of each signal by 2 or 3 methods of time domain analysis, frequency domain analysis, time-frequency domain analysis and neural network;
extracting the characteristic vector by time domain analysis, wherein the extracting of the characteristic vector comprises extracting a maximum value, a root mean square value, a waveform index, a peak index, a pulse index, a margin index, a kurtosis index and a skewness index of a signal;
the frequency domain analysis comprises the steps of converting the frequency domain into a frequency domain by adopting an analysis method of fast Fourier transform, and then extracting features of a mean square frequency domain, a root mean square frequency domain, a variance frequency and a standard deviation frequency;
the time-frequency domain analysis comprises the steps of extracting wavelet packet energy of each order of the signal by adopting wavelet transformation, and extracting IMF energy of each order of the signal by adopting empirical mode decomposition;
the neural network extracts the convolution features.
Further, the feature fusion of the normalized first feature vector includes:
Figure BDA0002418816050000031
wherein XiIs the ith first feature vector, X is the fused first feature vector, wiN is the first and second features including feature direction for the weighting factor weightThe total number of the quantities and the weight of the weighting factor satisfy:
Figure BDA0002418816050000032
further, the weighting factor weight is optimized by adopting a whale optimization algorithm to obtain the optimized weighting factor weight.
Further, optimizing the weighting factor weight by adopting a whale optimization algorithm comprises the following steps:
(1) using the vector formed by the weighting factors as a whale individual of the whale population; constructing a mathematical model surrounding the hunting behavior and a mathematical model of the hunting behavior;
the mathematical model surrounding the prey behavior is represented as follows:
D＝|CX*(t)-X(t)|
X(t+1)＝X*(t)-A·D
where t denotes the current number of iterations, A and C denote coefficients, X*(t) represents the optimal solution location vector so far, X (t) represents the location vector of the current search agent, D represents the distance of the current search agent location vector from the optimal solution location vector, A and C are given by the following equations:
A＝2a×r1-a
C＝2×r2
in the formula, r1And r2Is the random number in (0,1), the value of a decreases linearly from 2 to 0, a is 2-2 × T/Tmax，TmaxIn order to be the maximum number of iterations,
the mathematical model of the hunting behavior is as follows:
X(t+1)＝X*(t)+Dp·ebl·cos(2πl)
in the formula, Dp＝|X*(t) -X (t) l represents the distance between whale and prey, X*(t) represents the best position vector so far, b is a constant representing the shape of the spiral, and l is a random number in (-1, 1);
(2) calculating the fitness value f of the individual, and selecting the individual with the minimum fitness value and the position;
Figure BDA0002418816050000041
wherein, YiFor the ith predictor, Y, of the support vector machine diagnostic modeli' is the ith actual value of the training sample diagnosis result, and N is the total number of the training samples;
(3) randomly generating a random number P between (0,1), judging whether P is more than or equal to 0.5, if so, further judging whether | A | ≧ 1 is true, and if | A | ≧ 1, updating the next generation individual position by adopting the following method:
D＝|CXrand-X(t)|
X(t+1)＝Xrand-A·D
in the formula, XrandIs a randomly selected search agent location vector;
if | A | <1, then the next generation individual position is updated using the following equation:
X(t+1)＝X*(t)+Dp·ebl·cos(2πl)p≥0.5
if P <0.5, the next generation individual location is updated using the following equation:
X(t+1)＝X*(t)-A·D p＜0.5；
(4) judging whether the maximum iteration times is reached, if so, outputting an optimal fitness value and a position, wherein the position is the position for finally capturing food, namely the optimized weighting factor weight; otherwise, returning to the step (2).
Further, constructing the support vector machine model comprises:
diagnosing and identifying the real-time data through a support vector machine diagnosis model, wherein the identification result comprises a normal state and the fault type of a transmission system;
the fault types include gearbox faults, rotor faults, stator faults, bearing faults, base faults and unbalance/misalignment faults and transmission element faults.
Further, extracting the second feature vector includes: extracting the second feature vector comprises: and extracting second feature vectors corresponding to the first feature vectors one by one.
Further, the feature fusion of the normalized second feature vector includes: and carrying out weighted summation on the second feature vector by adopting the optimized weighting factor weight.
The technical scheme of the invention has the following beneficial technical effects:
(1) the invention collects multiple signals to extract the characteristics and perform characteristic fusion, can effectively overcome the problem of low accuracy caused by single characteristic diagnosis, and the obtained diagnosis result is more consistent with the real running state of the motor, thereby effectively improving the fault diagnosis rate of the transmission system.
(2) The fault diagnosis method provided by the invention realizes fault diagnosis of various fault factors, and the obtained fault diagnosis result is high in accuracy.
(3) According to the invention, whale algorithm is adopted to optimize the weighting factor weight of the feature fusion, so that the diagnosis accuracy is further improved. The whale optimization algorithm is simple to operate, few in adjusting parameters and only comprises two main adjusting parameters, the algorithm can well balance development and exploration capacity, and the overall optimization capacity is strong.
Drawings
FIG. 1 is a schematic diagram of a fault diagnosis process;
FIG. 2 is a schematic diagram of an adaptive weighted fusion architecture;
FIG. 3 is a schematic flow chart of an algorithm for optimizing whale;
FIG. 4 is a schematic diagram of a support vector machine;
FIG. 5 is an overall flow diagram of a transmission fault diagnosis.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings in conjunction with the following detailed description. It should be understood that the description is intended to be exemplary only, and is not intended to limit the scope of the present invention. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present invention.
For more and more complex conventional systems, a plurality of faults are correlated, and the fault characteristics are crossed. In order to improve the precision and reliability of fault diagnosis of a transmission system, the invention provides a transmission system fault diagnosis method based on multi-information fusion, which comprises the steps of collecting current, voltage signals, vibration signals, rotating speed signals, temperature and noise signals of the transmission system, preprocessing data, applying a related algorithm of feature extraction, extracting fault feature quantities to serve as a condition attribute set of fault classification, establishing a fault database, fusing the feature vectors by adopting a self-adaptive weighted fusion algorithm, establishing a diagnosis model by adopting a support vector machine model, collecting data of all feature quantities of the transmission system in real time, and diagnosing faults by a fault diagnosis method, so that fault diagnosis of various fault factors is realized, and the accuracy of the obtained fault diagnosis result is high.
With reference to fig. 1, the specific steps of the fault diagnosis method are as follows:
step 1, acquiring a plurality of groups of original signals of a transmission system in normal and fault states, wherein the original signals comprise current signals, vibration signals and 2-4 signals in voltage signals, rotating speed signals, temperature signals and noise signals; historical data of the drive train may be utilized. The data covers various fault conditions of the transmission system, and the integrity of the sample is further ensured.
Step 2, preprocessing the acquired time domain data by adopting preprocessing methods such as filtering, abnormal value processing and the like;
step 3, extracting the characteristic vector of each signal in the normal state and the fault state of the transmission system by adopting 2 or 3 methods of time domain analysis, frequency domain analysis, time-frequency domain analysis and neural network;
(1) time domain analysis:
the time domain analysis characteristic parameters comprise a maximum value, a root mean square value, a waveform index, a peak index, a pulse index, a margin index, a kurtosis index and a skewness index.
Maximum value:
Xmax＝max{|xi|}i＝1,2,…,N (1)
where N is the data length, xiTo measure the signal value.
Root mean square value:
Figure BDA0002418816050000071
the waveform index is as follows:
Figure BDA0002418816050000072
wherein
Figure BDA0002418816050000073
Is taken as the absolute average value of the average,
Figure BDA0002418816050000074
peak index:
Figure BDA0002418816050000075
pulse index:
Figure BDA0002418816050000076
margin indexes are as follows:
Figure BDA0002418816050000077
wherein XrIn order to obtain the square root amplitude value,
Figure BDA0002418816050000078
kurtosis index:
Figure BDA0002418816050000079
wherein
Figure BDA00024188160500000710
Distortion index:
Figure BDA00024188160500000711
wherein
Figure BDA0002418816050000081
(2) Frequency domain analysis:
the method comprises the steps of converting a signal into a frequency domain by adopting an analysis method of fast Fourier transform, and then extracting features of a mean square frequency domain, a root mean square frequency domain, a variance frequency and a standard deviation frequency.
(3) Time-frequency domain analysis:
the traditional Fourier transform is an integral transform, namely the representation of a signal is either completely in a time domain or completely in a frequency domain, and for a time-varying non-stationary signal, the change situation of a signal spectrum along with time is often expected, namely the time-frequency representation of the signal.
The wavelet analysis is a brand new signal time-scale analysis method, inherits the idea that Fourier analysis approaches any function by taking a simple harmonic function as a basic function, but the basic function of the wavelet analysis is a series of scale variable functions, so that the wavelet analysis has good time-frequency positioning characteristics and self-adaptive capacity to signals, and therefore, various time-varying signals can be effectively decomposed. The wavelet has the characteristic of multi-resolution analysis, and the instant frequency resolution is variable, so that the wavelet has the outstanding advantage of local analysis of non-stationary signals and has a good time-frequency positioning function.
The basic idea of wavelet transform is consistent with fourier transform, and a family of functions is also used to represent signals, and the family of functions is called wavelet function system, and can be scaled and translated to obtain a family of functions:
Figure BDA0002418816050000082
title psia,b(t) } is continuous wavelet, ψ is basic wavelet, where a is scale factor and b is translation factor. The scale factor changes the shape of the continuous wavelet and the shift factor changes the displacement of the wavelet.
Function f (t) e L with finite energy2(R) the continuous wavelet transform for ψ (t) is defined as
Figure BDA0002418816050000083
In the formula (I), the compound is shown in the specification,
Figure BDA0002418816050000084
denotes the complex conjugate, sign of<f,ψa,b>The inner product of the two is shown.
In practical application, discretization is required, and the wavelet function psi meeting the stable condition is used as the instruction
Figure BDA0002418816050000091
f at dimension 2jAnd the binary wavelet transform of x position is defined as
Figure BDA0002418816050000092
If the sampling value of the original signal is
Figure BDA0002418816050000093
Initial sampling value using discrete binary wavelet transform
Figure BDA0002418816050000094
Iterative solution, namely obtaining the approximation result of the original signal under different scales
Figure BDA0002418816050000095
Sum wavelet decomposition result
Figure BDA0002418816050000096
The iterative formula of the discrete wavelet is:
Figure BDA0002418816050000097
in the wavelet decomposition process, since only the low frequency part is decomposed again and the high frequency signal is not decomposed, the resolution of the high frequency band is poor, and the time resolution of the low frequency band is poor, so that the wavelet packet decomposition is generated at the same time. Different from wavelet decomposition, the decomposition of the low-frequency signals of each layer and the decomposition of the high-frequency signals are realized, so that a more precise analysis method can be provided for the signals, and the time-frequency resolution of the signals is improved.
Wavelet packet decomposition algorithm consists of
Figure BDA0002418816050000098
To find
Figure BDA0002418816050000099
And
Figure BDA00024188160500000910
the formula of (1) is:
Figure BDA00024188160500000911
Figure BDA00024188160500000912
reconstructing signals after decomposing the wavelet packet, selecting a dual filter of the wavelet packet filter in the decomposing process, reconstructing the signals of each sub-node space,
wavelet packet reconstruction algorithm
Figure BDA00024188160500000913
And
Figure BDA00024188160500000914
to find
Figure BDA00024188160500000915
The formula of (1) is:
Figure BDA00024188160500000916
in the formula, pk、qkAre respectively hk、gkThe dual filter of (1).
When a single node signal is reconstructed, the wavelet packet coefficients of other sub-nodes are set to be zero to obtain a time domain signal only containing the frequency band information, then the reconstructed energy of each node is calculated, and the energy of each node of the wavelet packet is normalized to obtain a normalized wavelet packet scale energy spectrum. Wavelet packet energy of each order of the signal is extracted by adopting wavelet transformation, and IMF energy of each order of the signal is extracted by adopting empirical mode decomposition.
The empirical mode decomposition is a novel self-adaptive signal time-frequency processing method, is particularly suitable for analysis processing of nonlinear non-stationary signals, and has the advantages that the method processes the signals according to the self time scale characteristics without setting any basis function, which is essentially different from the Fourier decomposition and the wavelet decomposition which are established on the prior harmonic basis function and wavelet basis function.
The core of the method is empirical mode decomposition, which can decompose a complex signal into a finite number of eigenmode functions (IMFs), each decomposed IMF component comprises local characteristic signals of different time scales of an original signal, and the decomposition is based on the local characteristics of the time scales of a signal sequence, so that the method has self-adaptability. The eigenmode function must have 2 conditions met:
condition 1: in the whole time range of the function, the number of local extreme points and zero-crossing points must be equal, or the difference is at most 1;
condition 2: at any point in time, the local maximum envelope and the local minimum envelope must be 0 on average.
For a given signal, an EMD method is adopted to solve an intrinsic mode function, and the screening process of EMD analysis is as follows:
(a) finding out all extreme points on x (t), and connecting all extreme points and minimum points with a curve to form an upper envelope fmax(t) and lower envelope fmin(t), the mean value of the upper envelope line and the lower envelope line is m (t).
(b) The original signals x (t) and m (t) are subtracted, and the difference is recorded as h1(t), there is the following equation:
h1(t)＝x(t)-m(t) (17)
ideally, h1(t) is a fundamental mode component, whereas the actual signal is generally more complex, h1(t) repeating the above process as new x (t) until the condition is satisfied, then h1(t) is the first fundamental mode component, noted:
c1(t)＝h1(t) (18)
(c) first fundamental mode component c1(t) after decomposition, subtracting c from the original signal x (t)1(t) obtaining a residual signal x1(t)：
x1(t)＝x(t)-c1(t) (19)
(d) X is to be1(t) as an original signal x (t), repeating the steps (1) to (3) to obtain a second IMF component c in sequence2(t)，Third IMF component c3(t) up to the nth IMF component cnAnd (t), finally decomposing the original signal into n IMF components after EMD decomposition.
Extracting time domain features of IMF components:
(a) determining the number n of IMF components, and performing EMD on the acquired signals to obtain n-order IMF components;
(b) the energy of the first n-order IMF component is calculated,
(c) the total energy of all IMFs was calculated.
And subsequently, calculating normalized energy for the total energy of the IMF, and taking the normalized energy distribution as a fault feature vector.
(4) And (5) constructing a neural network model, training and extracting convolution characteristics.
Step 4, normalizing the feature vectors extracted in the step 3 under the normal and fault states;
step 5, performing feature fusion on the normalized feature vector obtained in the step 4 by applying a self-adaptive weighting algorithm;
the fusion process of the self-adaptive weighted fusion algorithm is as follows: and (3) respectively multiplying each feature vector by a corresponding weight through the extracted feature vector of each parameter, and finally adding to obtain the fused features, wherein the feature fusion process is combined with the graph 2. Wherein, X1, X2, X3, …, Xn respectively represent the feature vectors extracted from the data collected by each channel, w1, w2, w3, …, wn respectively represent the weights corresponding to each channel, Σ represents the summation processing, and X represents the vector finally obtained by fusion. The final feature vector obtained by feature weighting can be expressed as follows:
Figure BDA0002418816050000111
in addition, the mathematical relationship of the weighting factor weights is expressed as follows:
Figure BDA0002418816050000112
further, the adaptive weighting algorithm in the step 5 comprises optimizing the weights by adopting a whale optimization algorithm.
Because the weights in the traditional weighted fusion algorithm are generally set to be random or 1/n, the weights are averaged, that is, the weights of each channel are the same, so that the feature vector obtained after weighted fusion does not accord with the actual fault condition, and finally the diagnostic result of the model is not ideal. In addition, because the contribution of the data acquired by each channel to the final fault identification is different, finding the optimal weight corresponding to each channel is a fundamental method for solving the problem, the invention adopts a colony intelligent algorithm, namely a whale optimization algorithm, to optimize the weight in the feature fusion process, finds the optimal weight, finally enables the feature vector obtained by fusion to better accord with the actual working state of a transmission system, and further enables the fault diagnosis identification rate to be more ideal.
In the whale algorithm, the position of each individual head whale represents a candidate solution for optimizing the problem in the feasible domain, and is referred to as a search agent. However, there is no known a priori knowledge in solving the global optimization problem, i.e., no information of the target prey is determined, so the search agent with the lowest population fitness value is taken as the target prey. Other search agents of the population then iteratively update their own location based on the location of the best agent.
The whale optimization algorithm is a novel colony intelligent optimization algorithm, the whole process of the algorithm is to provide mathematical models of all stages according to different stages of predation in the whole process of simulating the predation of the whale. The whole predation process of whales can be divided into three aspects of hunting, hunting behavior and hunting.
In the method, weights corresponding to all channels are combined into a vector to serve as an individual of a whale population, namely a search agent, a limited number of whale individuals combined by the weights are initialized in a space, through continuous iterative updating, the whale individuals are enabled to be close to food continuously, and finally the food is captured, and the individual of the captured food is an optimal solution of the weights. With reference to fig. 3, the process of optimizing the weighting factor weight includes:
(1) taking a vector formed by the weighting factors as a whale individual of the whale population, constructing a mathematical model, initializing, and initializing N whale individuals in space as an initialized population; the mathematical models include a mathematical model encompassing the behavior of a hunting, a mathematical model of a hunting behavior, and a mathematical model of a search for a hunting.
The mathematical model surrounding the prey behavior is represented as follows:
D＝|CX*(t)-X(t)| (22)
X(t+1)＝X*(t)-A·D (23)
where t denotes the current number of iterations, A and C denote coefficients, X*(t) represents the best solution location vector so far, X (t) represents the location vector of the current search agent, A and C are given by the following equations:
A＝2a×r1-a (24)
C＝2×r2 (25)
in the formula, r1And r2Is a random number in (0,1), the value of a decreases linearly from 2 to 0, T represents the current number of iterations, TmaxIs the maximum number of iterations.
a＝2-2×t/Tmax (26)
According to the hunting behavior of whale, which swims to the prey in spiral motion, the mathematical model of the hunting behavior is as follows:
X(t+1)＝X*(t)+Dp·ebl·cos(2πl) (27)
in the formula, Dp＝|X*(t) -X (t) l represents the distance between whale and prey, X*(t) represents the best position vector so far, b is a constant that defines the shape of the spiral, and l is a random number in (-1, 1).
Notably, whales swim to the prey in a spiral shape while contracting the enclosure. Therefore, in the synchronous behavior model, there are a probability of 0.5 for selecting the contraction and wrapping mechanism and selecting the spiral model to update the position of whale at the next moment, respectively, and the mathematical model is as follows:
X(t+1)＝X*(t)-A·D p＜0.5 (28)
X(t+1)＝X*(t)+Dp·ebl·cos(2πl)p≥0.5 (29)
when the prey is attacked, a value for decreasing a is set close to the prey on the mathematical model, so that the fluctuation range of A also decreases with a. In an iterative process when the value of a falls from 2 to 0, A is a random value within [ -a, a ], when the value of A is within [ -1,1], the next position of the whale can be any position between its present position and the position of the prey, and the algorithm sets that when A <1, the whale makes an attack on the prey.
In order to find a better prey location, the whale in the sitting position needs to search other prey randomly within the search range in addition to the above burble-net hunting strategy, and the mathematical model of the searched prey is represented as follows:
D＝|CXrand-X(t)| (30)
X(t+1)＝Xrand-A·D (31)
in the formula, XrandIs a randomly selected search agent location vector.
(2) Calculating the fitness value f of the individual, and selecting the individual with the minimum fitness value and the position;
Figure BDA0002418816050000141
wherein, YiFor the ith predictor, Y, of the support vector machine diagnostic modeli' is the ith actual value of the training sample diagnosis result, and N is the total number of the training samples;
(3) randomly generating a random number P between (0,1), judging whether P is more than or equal to 0.5, if so, further judging whether | A | > or equal to 1 is true, and if | A | > or equal to 1, updating the next generation individual position by adopting a formula (30) (31); if the | A | <1, updating the next generation individual position by adopting a formula (29); if P <0.5, the next generation individual location is updated using equation (28).
In order to ensure exploration and convergence, the algorithm is set to randomly select a search agent when A is larger than or equal to 1, and the positions of other whales are updated according to the randomly selected whale positions, so that the whales are forced to deviate from prey, and a more appropriate prey is found, thereby enhancing the exploration capacity of the algorithm and enabling the WOA algorithm to carry out global search.
(4) Judging whether the maximum iteration times is reached, if so, outputting an optimal fitness value and a position, wherein the position is the position for finally capturing food, namely the optimized weighting factor weight; otherwise, returning to the step (2).
Step 6, training a support vector machine model by using the feature fusion vector obtained in the step 5 to obtain a support vector machine diagnosis model;
a Support Vector Machine (SVM) follows the principle of minimizing structural risks, seeks a global optimal solution, integrates methods such as a linear learning machine, a kernel function theory, an optimization theory and the like, and theoretically well overcomes the problems inevitably generated by the traditional learning algorithm such as dimension disaster, overfitting, local extremum and the like. And can achieve good classification capability under the condition of few training samples. The SVM is firstly analyzed aiming at the linear divisible condition, and for the linear inseparable condition, a mapping algorithm is used for converting low-dimensional linear inseparable samples in an input space into a high-dimensional feature space so as to enable the samples to be linearly divisible, so that a linear algorithm can be adopted in the high-dimensional feature space, and the analysis of the nonlinear features of the samples is realized.
The SVM is developed from an optimal classification surface under a linear separable condition, and the basic idea can be represented by FIG. 4, wherein solid points and hollow points in FIG. 4 represent two different types of samples respectively, H is a classification line, H is1And H2The distance between the two straight lines which are respectively the nearest sample to the classification line and are parallel to the classification line is called the classification interval. What is called an optimal classification line is that the classification line not only can correctly separate two classes, but also can maximize the classification interval.
The classification line equation can be expressed as (w · x) + b ═ 0, where w is the normal vector and b is the offset. w, b can satisfy the hyperplane of the following constraints:
Figure BDA0002418816050000151
the constraint defines the norm of w as the inverse of its distance from the nearest point of the hyperplane.
x to the hyperplane (w, b) is a distance d (w, b; x)
Figure BDA0002418816050000152
The classification interval is then expressed as follows:
Figure BDA0002418816050000153
thus, the hyperplane that minimizes the following is the optimal hyperplane:
Figure BDA0002418816050000154
in summary, the problem of finding the optimal hyperplane is to solve the quadratic optimization problem constrained by the following inequality:
Figure BDA0002418816050000161
according to the optimization theory, the Lagrange multiplier method can be used for converting the optimization problem into the following unconstrained optimization problem:
Figure BDA0002418816050000162
s.t.yi[(w·x)+b]≥1,i＝1,2,…,n
according to the unconstrained theory, the minimum value of the Lagrange function is calculated for w and b to obtain:
Figure BDA0002418816050000163
based on Lagrange dual principle, without constraint
Figure BDA0002418816050000164
Is equivalent to solving the dual problem thereof,
Figure BDA0002418816050000165
will be provided with
Figure BDA0002418816050000166
The following dual optimization problem is obtained by introducing the extreme value condition:
Figure BDA0002418816050000167
Figure BDA0002418816050000168
the solution of the problem is
Figure BDA0002418816050000169
Thus, an optimal hyperplane is obtained, namely, the segmentation of different types of faults is as follows:
Figure BDA0002418816050000171
b*:yi{(w*·xi)+b*}＝1
step 7, collecting 3 to 5 signals in current, voltage signals, vibration signals, rotating speed signals, temperature and noise signals of the transmission system in real time; the type of signal collected is the same as in step 1.
And 8, preprocessing each signal acquired in real time, including filtering and abnormal value processing.
Step 9, respectively extracting second feature vectors and normalizing the second feature vectors; and extracting second feature vectors corresponding to the first feature vectors one by adopting a time domain analysis method, a frequency domain analysis method, a time-frequency domain analysis method and a neural network method. The way of extracting the feature vector is the same as step 3.
Step 10, performing feature fusion on each normalized second feature vector; and 5, carrying out weighted summation on the second feature vector by adopting the optimized weighting factor weight obtained in the step 5.
And 11, diagnosing the fused feature vector by adopting the trained support vector machine diagnosis model, and outputting a fault detection result.
In summary, the invention relates to a transmission system fault diagnosis method based on multi-information fusion, which obtains a plurality of groups of current, voltage signals, vibration signals, rotating speed signals, temperature and noise signals of a transmission system in normal and fault states; preprocessing each acquired signal; respectively extracting first feature vectors, and performing normalization processing and feature fusion to obtain training samples; constructing a support vector machine model, and training by using a training sample; and (3) acquiring each signal of the transmission system in real time for preprocessing, extracting a second feature vector and normalizing, diagnosing by adopting the trained support vector machine diagnosis model after feature fusion, and outputting a fault detection result. The invention collects multiple signals to extract the characteristics and perform characteristic fusion, can effectively overcome the problem of low accuracy rate caused by diagnosis by single characteristic, obtains a diagnosis result which is more consistent with the real running state of the motor, and effectively improves the fault diagnosis rate of the transmission system.
It is to be understood that the above-described embodiments of the present invention are merely illustrative of or explaining the principles of the invention and are not to be construed as limiting the invention. Therefore, any modification, equivalent replacement, improvement and the like made without departing from the spirit and scope of the present invention should be included in the protection scope of the present invention. Further, it is intended that the appended claims cover all such variations and modifications as fall within the scope and boundaries of the appended claims or the equivalents of such scope and boundaries.

Claims (6)

1. A transmission system fault diagnosis method based on multi-information fusion is characterized by comprising the following steps:
acquiring a plurality of groups of original signals of the transmission system in normal and fault states, wherein the original signals comprise current signals, vibration signals and 2-4 signals in voltage signals, rotating speed signals, temperature signals and noise signals; preprocessing the acquired original signal; respectively extracting first feature vectors; normalizing each first feature vector; performing feature fusion on the normalized first feature vector by adopting a self-adaptive weighting algorithm, and taking the fused first feature as a training sample;
constructing a support vector machine model, and training the support vector machine model by using a training sample to output a diagnosis result;
acquiring a current signal, a vibration signal and the 2 to 3 signals of the transmission system in real time; preprocessing each signal acquired in real time, respectively extracting and normalizing second feature vectors, and performing feature fusion on each normalized second feature vector; diagnosing the fused feature vector by adopting the trained support vector machine diagnosis model, and outputting a fault detection result;
the feature fusion of the normalized first feature vector comprises the following steps:
Figure FDA0003231041340000011
wherein XiIs the ith first feature vector, X is the fused first feature vector, wiN is the total number of the first and second features including the feature vector, and the weighting factor weight satisfies:
Figure FDA0003231041340000012
optimizing the weighting factor weight by adopting a whale optimization algorithm to obtain an optimized weighting factor weight;
the optimization of the weighting factor weight by adopting a whale optimization algorithm comprises the following steps:
(1) using the vector formed by the weighting factors as a whale individual of the whale population; constructing a mathematical model surrounding the hunting behavior and a mathematical model of the hunting behavior;
the mathematical model surrounding the prey behavior is represented as follows:
D＝|CX*(t)-X(t)|
X(t+1)＝X*(t)-A·D
where t denotes the current number of iterations, A and C denote coefficients, X*(t) represents the optimal solution location vector so far, X (t) represents the current search agent, D represents the distance of the current search agent location vector from the optimal solution location vector, A and C are given by the following equations:
A＝2a×r1-a
C＝2×r2
in the formula, r1And r2Is the random number in (0,1), the value of a decreases linearly from 2 to 0, a is 2-2 × T/Tmax，TmaxIs the maximum iteration number;
the mathematical model of the hunting behavior is as follows:
X(t+1)＝X*(t)+Dp·ebl·cos(2πl)
in the formula, Dp＝|X*(t) -X (t) l represents the distance between whale and prey, X*(t) represents the best position vector so far, b is a constant representing the shape of the spiral, and l is a random number in (-1, 1);
(2) calculating the fitness value f of the individual, and selecting the individual with the minimum fitness value and the position;
Figure FDA0003231041340000021
wherein, YiFor the ith predictor, Y, of the support vector machine diagnostic modeli' is the ith actual value of the training sample diagnosis result, and N is the total number of the training samples;
(3) randomly generating a random number P between (0,1), judging whether P is more than or equal to 0.5, if so, further judging whether | A | ≧ 1 is true, and if | A | ≧ 1, updating the next generation individual position by adopting the following method:
D＝|CXrand-X(t)|
X(t+1)＝Xrand-A·D
in the formula, XrandIs a randomly selected search agent location vector;
if | A | <1, then the next generation individual location is updated using the following equation:
X(t+1)＝X*(t)+Dp·ebl·cos(2πl)p≥0.5
if P <0.5, the next generation individual location is updated using the following equation:
X(t+1)＝X*(t)-A·D p＜0.5；
(4) judging whether the maximum iteration times is reached, if so, outputting an optimal fitness value and a position, wherein the position is the position for finally capturing food, namely the optimized weighting factor weight; otherwise, returning to the step (2).
2. The multi-information fusion-based transmission system fault diagnosis method according to claim 1, wherein the preprocessing of the acquired signals and the preprocessing of the signals acquired in real time comprise filtering and abnormal value processing.
3. The multi-information fusion based transmission system fault diagnosis method according to claim 1 or 2, wherein extracting the first feature vector comprises extracting the feature vector of each signal by 2 or 3 methods of time domain analysis, frequency domain analysis, time-frequency domain analysis and neural network;
extracting the characteristic vector by time domain analysis, wherein the extracting of the characteristic vector comprises extracting a maximum value, a root mean square value, a waveform index, a peak index, a pulse index, a margin index, a kurtosis index and a skewness index of a signal;
the frequency domain analysis comprises the steps of converting a signal into a frequency domain by adopting an analysis method of fast Fourier transform, and then extracting features of a mean square frequency domain, a root mean square frequency domain, a variance frequency and a standard deviation frequency;
the time-frequency domain analysis comprises the steps of extracting wavelet packet energy of each order of the signal by adopting wavelet transformation, and extracting IMF energy of each order of the signal by adopting empirical mode decomposition;
the neural network extracts the convolution features.
4. The multi-information fusion based transmission system fault diagnosis method according to claim 1, wherein constructing a support vector machine model comprises:
diagnosing and identifying the real-time data through a support vector machine diagnosis model, wherein the identification result comprises a normal state and the fault type of a transmission system;
the fault types include gearbox faults, rotor faults, stator faults, bearing faults, pedestal faults, and unbalance/misalignment faults, and transmission element faults.
5. The multi-information fusion based transmission system fault diagnosis method according to claim 4, wherein extracting the second feature vector comprises: and extracting second feature vectors corresponding to the first feature vectors one by one.
6. The multi-information fusion based transmission system fault diagnosis method according to claim 4, wherein the feature fusion of the normalized second feature vector comprises: and carrying out weighted summation on the second feature vector by adopting the optimized weighting factor weight.
CN202010199373.1A 2020-03-20 2020-03-20 Transmission system fault diagnosis method based on multi-information fusion Active CN111307453B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010199373.1A CN111307453B (en) 2020-03-20 2020-03-20 Transmission system fault diagnosis method based on multi-information fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010199373.1A CN111307453B (en) 2020-03-20 2020-03-20 Transmission system fault diagnosis method based on multi-information fusion

Publications (2)

Publication Number Publication Date
CN111307453A CN111307453A (en) 2020-06-19
CN111307453B true CN111307453B (en) 2021-11-12

Family

ID=71151452

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010199373.1A Active CN111307453B (en) 2020-03-20 2020-03-20 Transmission system fault diagnosis method based on multi-information fusion

Country Status (1)

Country Link
CN (1) CN111307453B (en)

Families Citing this family (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111929044B (en) * 2020-07-15 2023-08-08 西门子工厂自动化工程有限公司 Method, apparatus, computing device and storage medium for monitoring device status
CN111947928B (en) * 2020-08-10 2021-11-05 山东大学 Multi-source information fusion bearing fault prediction system and method
CN112052796A (en) * 2020-09-07 2020-12-08 电子科技大学 Permanent magnet synchronous motor fault diagnosis method based on deep learning
CN112326241B (en) * 2020-09-11 2023-10-13 西安英特迈思信息科技有限公司 Nuclear power main pump bearing fault early warning method based on fusion degradation index
CN112613157A (en) * 2020-11-26 2021-04-06 北京航天智造科技发展有限公司 Rotor fault analysis method and device
CN112633147A (en) * 2020-12-22 2021-04-09 电子科技大学 Magnetic anomaly detection method based on multi-feature fusion and intelligent iterative optimization algorithm
CN113720376A (en) * 2020-12-29 2021-11-30 宇力源(深圳)科技有限公司 Modular crawler with system self-checking function and self-checking method thereof
CN113253050B (en) * 2021-05-08 2022-07-12 南昌大学 Traveling wave fault location method based on whale optimization Kalman filtering algorithm
CN113487141B (en) * 2021-06-11 2023-09-29 北京控制工程研究所 Bearing state evaluation method based on multisource information clustering fusion
CN113325314A (en) * 2021-07-20 2021-08-31 安徽巨森电器有限公司 Motor fault diagnosis method
CN113970440A (en) * 2021-10-22 2022-01-25 南京高精齿轮集团有限公司 Method for judging abnormal sound of gear box, diagnosis system and storage medium
CN114252261B (en) * 2021-12-21 2024-03-08 沈阳顺义科技股份有限公司 Fault diagnosis method and system for steering system of comprehensive transmission device
CN114943259B (en) * 2022-07-22 2022-10-21 深圳市世坤科技实业有限公司 Equipment fault diagnosis system and method for processing electric element by laser
CN116150677A (en) * 2023-04-19 2023-05-23 国家石油天然气管网集团有限公司 Support vector machine-based oil transfer pump fault early warning method and system
CN117150414B (en) * 2023-10-17 2024-04-12 广东迅扬科技股份有限公司 Fault diagnosis method
CN117556344A (en) * 2024-01-08 2024-02-13 浙江大学 Fault diagnosis method and system for ball mill transmission system based on multi-source information fusion

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2405824A1 (en) * 2000-04-11 2001-10-18 Telstra New Wave Pty Ltd A gradient based training method for a support vector machine
WO2006043308A1 (en) * 2004-10-18 2006-04-27 Fujitsu Limited Operation management program, operation management method, and operation management device
CN102520341A (en) * 2011-12-05 2012-06-27 南京航空航天大学 Analog circuit fault diagnosis method based on Bayes-KFCM (Kernelized Fuzzy C-Means) algorithm
CN103412220A (en) * 2013-08-13 2013-11-27 无锡科技职业学院 Ship power station fault diagnosis method based on data fusion
CN105372087A (en) * 2015-11-30 2016-03-02 南通大学 Fault diagnosis method based on multi-sensor signal analysis
CN105911975A (en) * 2016-04-14 2016-08-31 南京航空航天大学 Integrated diagnosis method of gas path fault of aeroengine
CN107590506A (en) * 2017-08-17 2018-01-16 北京航空航天大学 A kind of complex device method for diagnosing faults of feature based processing
CN108710889A (en) * 2018-04-02 2018-10-26 天津大学 A kind of scarce cylinder method for diagnosing faults of automobile engine

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2405824A1 (en) * 2000-04-11 2001-10-18 Telstra New Wave Pty Ltd A gradient based training method for a support vector machine
WO2006043308A1 (en) * 2004-10-18 2006-04-27 Fujitsu Limited Operation management program, operation management method, and operation management device
CN102520341A (en) * 2011-12-05 2012-06-27 南京航空航天大学 Analog circuit fault diagnosis method based on Bayes-KFCM (Kernelized Fuzzy C-Means) algorithm
CN103412220A (en) * 2013-08-13 2013-11-27 无锡科技职业学院 Ship power station fault diagnosis method based on data fusion
CN105372087A (en) * 2015-11-30 2016-03-02 南通大学 Fault diagnosis method based on multi-sensor signal analysis
CN105911975A (en) * 2016-04-14 2016-08-31 南京航空航天大学 Integrated diagnosis method of gas path fault of aeroengine
CN107590506A (en) * 2017-08-17 2018-01-16 北京航空航天大学 A kind of complex device method for diagnosing faults of feature based processing
CN108710889A (en) * 2018-04-02 2018-10-26 天津大学 A kind of scarce cylinder method for diagnosing faults of automobile engine

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Comparison between Genetic Algorithm and Whale Optimization Algorithm in Fault Location Estimation in Power Systems;A.Sanad Ahmed 等;《2017 Nineteenth International Middle East Power Systems Conference (MEPCON)》;20180227;全文 *
EMD分解与多特征融合的齿轮故障诊断方法;秦波 等;《机床与液压》;20160215;第44卷(第3期);全文 *
基于深度学习特征提取和WOA-SVM状态识别的轴承故障诊断;赵春华 等;《振动与冲击》;20190528;第38卷(第10期);全文 *

Also Published As

Publication number Publication date
CN111307453A (en) 2020-06-19

Similar Documents

Publication Publication Date Title
CN111307453B (en) Transmission system fault diagnosis method based on multi-information fusion
CN112308158B (en) Multi-source field self-adaptive model and method based on partial feature alignment
Sun et al. An experimental study on hyper-parameter optimization for stacked auto-encoders
CN112364779A (en) Underwater sound target identification method based on signal processing and deep-shallow network multi-model fusion
CN109410238B (en) Wolfberry identification and counting method based on PointNet + + network
CN112001306A (en) Electroencephalogram signal decoding method for generating neural network based on deep convolution countermeasure
CN113693563B (en) Brain function network classification method based on hypergraph attention network
CN101196905A (en) Intelligent pattern searching method
CN114282571B (en) Method, system, equipment and medium for constructing multidimensional health index of bearing
CN114169110B (en) Motor bearing fault diagnosis method based on feature optimization and GWAA-XGboost
CN111650453A (en) Power equipment diagnosis method and system based on windowing characteristic Hilbert imaging
CN113673346A (en) Motor vibration data processing and state recognition method based on multi-scale SE-Resnet
Lee et al. An adaptive resource-allocating network for automated detection, segmentation, and classification of breast cancer nuclei topic area: image processing and recognition
CN115130514A (en) Construction method and system for health index of engineering equipment
CN113208641B (en) Auxiliary diagnosis method for lung nodule based on three-dimensional multi-resolution attention capsule network
CN109190505A (en) The image-recognizing method that view-based access control model understands
CN115909016A (en) System, method, electronic device, and medium for analyzing fMRI image based on GCN
CN109165586A (en) intelligent image processing method for AI chip
CN113295410B (en) Bearing fault diagnosis method under variable rotating speed working condition
CN109165587A (en) intelligent image information extraction method
Yao et al. A hierarchical adversarial multi-target domain adaptation for gear fault diagnosis under variable working condition based on raw acoustic signal
CN116228916B (en) Image metal artifact removal method, system and equipment
CN116228915B (en) Image reconstruction method, system and equipment based on region judgment
Sun et al. IM-Net: Semantic Segmentation Algorithm for Medical Images Based on Mutual Information Maximization
Su et al. Multidimensional Time Series Prediction Based on WOA-BP-A Case Study of Wordle

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant