CN112069918A

CN112069918A - Fault diagnosis method and device for planetary gearbox

Info

Publication number: CN112069918A
Application number: CN202010823453.XA
Authority: CN
Inventors: 郑坤鹏; 丁云飞
Original assignee: Shanghai Dianji University
Current assignee: Shanghai Dianji University
Priority date: 2020-08-17
Filing date: 2020-08-17
Publication date: 2020-12-11

Abstract

The invention relates to a fault diagnosis method and a fault diagnosis device for a planetary gearbox, wherein the method comprises the following steps of: s1: acquiring a vibration signal of the planetary gearbox; s2: decomposing the vibration signal by adopting an MEEMD decomposition algorithm to obtain an IMF function; s3: calculating sample entropy according to the IMF function, and extracting fault characteristics of the planetary gearbox; s4: and according to the fault characteristics, carrying out classification and identification on fault types by adopting a WOA-LSSVM classification model. Compared with the prior art, the invention adopts the MEEMD decomposition algorithm, which not only inhibits the modal aliasing phenomenon, but also limits the defects of the EEMD algorithm; the regularization parameters and kernel function parameters of the LSSVM model are optimized by using a whale optimization algorithm, the method has the characteristics of simple and quick operation, strong global search capability and the like, and the local extreme value can be eliminated with great probability.

Description

Fault diagnosis method and device for planetary gearbox

Technical Field

The invention relates to the field of fault diagnosis of mechanical equipment, in particular to a fault diagnosis method and device of a planetary gear box.

Background

With the continuous advance of industrialization, the planetary gear box is applied to the transmission system of more and more rotating machines, and accurate fault diagnosis of the planetary gear box is crucial to the stable operation of the rotating machines. The traditional method for processing the vibration signal of the planetary gearbox is mostly wavelet analysis method, Empirical Mode Decomposition (EMD) method and the like. After the vibration signal of the planetary gearbox is processed, the fault characteristics of the planetary gearbox need to be extracted, generally time domain characteristics, and the parameters are more and difficult to select.

With the development of artificial intelligence technology, it is a trend to implement fault diagnosis of machinery by using an intelligent algorithm, and classical intelligent classification algorithms include a neural network, a support vector machine, a decision tree and the like, wherein the support vector machine is most widely applied due to the characteristics of simple algorithm, few required samples, excellent classification result and the like.

Compared with a wavelet analysis method, the Empirical Mode Decomposition (EMD) has certain adaptivity, but still has the problem of mode aliasing, and the Ensemble EMD (EEMD) has a certain inhibition effect on the mode aliasing by adding white noise to an original signal. However, the number of iterations of the EEMD algorithm is large, the calculation amount is increased, and if the added white noise is not appropriate, the decomposed components can not necessarily meet the IMF definition, so that more meaningless IMF components appear. The extraction of subsequent planetary gearbox faults is not facilitated. The time domain fault characteristics of the vibration signal are more and difficult to select.

Although the support vector machine algorithm can achieve an excellent classification effect, the calculation efficiency is low, the regularization parameter gamma and the kernel function parameter sigma are difficult to select, and if the selected parameters are not appropriate, the classification precision is reduced.

Disclosure of Invention

The invention aims to overcome the defects that the prior art ensemble empirical mode decomposition method has more meaningless IMF components and influences the fault diagnosis accuracy, and provides a fault diagnosis method and device of a planetary gearbox.

The purpose of the invention can be realized by the following technical scheme:

a method of fault diagnosis of a planetary gearbox comprising the steps of:

s1: acquiring a vibration signal of the planetary gearbox;

s2: decomposing the vibration signal by adopting an MEEMD decomposition algorithm to obtain an IMF function;

s3: extracting fault features of the planetary gearbox according to the IMF function;

s4: and according to the fault characteristics, carrying out classification and identification on fault types.

Further, the MEEMD decomposition algorithm comprises the following steps:

s201: adding a group of white noise signals with amplitudes opposite to each other and the mean value of the white noise signals is zero into the vibration signals respectively to obtain a first processing signal and a second processing signal respectively;

s202: performing EMD on the first processing signal and the second processing signal respectively to obtain multi-order IMF components;

s203: performing integrated averaging on the jth-order IMF component in the first processing signal and the jth-order IMF component in the second processing signal to obtain a jth-order integrated average component;

s204: calculating the permutation entropy of the integrated average components;

s205: judging whether the permutation entropy calculated in the step S204 is greater than a preset threshold, if so, executing a step S206, otherwise, executing a step S207;

s206: marking the j-th integrated average component as an abnormal signal, executing j ═ j +1, and returning to the step S203;

s207: and removing abnormal signals from the vibration signals acquired in the step S201 to obtain residual signals, and performing EMD decomposition on the residual signals to acquire the IMF function.

Further, the amplitude of the white noise signal is within a range of 0.1 to 0.2 times the standard deviation of the vibration signal.

Further, the calculation expression of the integrated average component of the j-th order is as follows:

in the formula I_j(t) is the integrated average component of the j-th order, N is the length of the vibration signal, Ne is the logarithm of the white noise signal,

the IMF component of the j-th order in the first processed signal,

the IMF component of the j-th order in the second processed signal.

Further, in step S4, a previously established and trained LSSVM classification model is used to perform classification and identification of the fault type.

Further, in the step S4, a WOA-LSSVM classification model which is pre-established and trained is adopted to perform classification and identification of the fault types, and the WOA-LSSVM classification model is an LSSVM classification model which optimizes regularization parameters and kernel function parameters by using a whale optimization algorithm.

Further, the training process of the WOA-LSSVM classification model comprises the following steps:

s401: determining a fitness function, and acquiring a training sample of a vibration signal of the planetary gearbox;

s402: initializing parameters of the whale optimization algorithm and the LSSVM classification model;

s403: loading the training samples into the LSSVM classification model for model training, and calculating a fitness function of the LSSVM classification model;

s404: based on the self-adaptive function, updating regularization parameters and kernel function parameters of the LSSVM classification model by adopting the whale optimization algorithm;

s405: circularly executing the step S403 to the step S404 until reaching the preset maximum iteration number;

s406: and establishing the WOA-LSSVM classification model by adopting the optimal fitness function calculation value in the iteration process and the corresponding regularization parameter and kernel function parameter.

Further, the fitness function is the classification accuracy of the training samples.

Further, in the step S4, the fault types identified by the classification include normal, broken tooth, wear and pitting.

The invention also provides a fault diagnosis device of the planetary gearbox, which comprises a memory and a processor, wherein the memory stores a computer program, and the processor calls the computer program to execute the steps of the method.

Compared with the prior art, the invention has the following advantages:

(1) according to the invention, the MEEMD decomposition algorithm is adopted to extract fault characteristics, and the interference of white noise signals is reduced and the completeness of the EEMD algorithm is improved by adding the white noise signals with opposite signs in pairs; by calculating the permutation entropy of the IMF components, separating abnormal signals of the time sequence signals according to the permutation entropy value and a certain rule, and then performing EMD decomposition on the residual signals, the modal aliasing phenomenon is inhibited, and the defects of the EEMD algorithm are limited.

(2) According to the invention, the LSSVM classification model is adopted to classify and identify the fault types, and compared with the traditional support vector machine, the objective function is changed from inequality constraint to equality constraint, so that the complexity of calculation is greatly simplified; experiments prove that the LSSVM algorithm obtains a good compromise between the complexity of the model and the accuracy of a training result, and is more suitable for the situation that a data sample is complex.

(3) The method considers that some classical intelligent optimization algorithms are often used for optimizing regularization parameters and kernel function parameters of the LSSVM model, such as particle swarm optimization algorithm, genetic algorithm and the like; although the algorithm has a certain improvement on the classification effect of the LSSVM, the trap of a local extreme value is not easy to jump out, so that the classification precision is not high; the regularization parameters and kernel function parameters of the LSSVM model are optimized by adopting a whale optimization algorithm, and compared with the algorithms, the algorithm has the characteristics of simple and quick operation, strong global search capability and the like, and has great probability of getting rid of local extreme values.

(4) According to the invention, the MEEMD algorithm is adopted to decompose the vibration signal, compared with the traditional EEMD algorithm, the modal aliasing is inhibited, the problem of excessive meaningless components is solved, a better decomposition effect is obtained, and a foundation is laid for the subsequent gearbox fault extraction; the method has the advantages that the MEEMD algorithm and the sample entropy are used for constructing the fault feature vector, and compared with the traditional time domain feature, the method is simple and convenient; and finally, aiming at the difficult problem of LSSVM parameter selection, optimizing the LSSVM by using a Whale Optimization Algorithm (WOA), establishing a WOA-LSSVM planetary gearbox fault diagnosis model, and realizing fault diagnosis of the gearbox.

Drawings

FIG. 1 is a MEEMD decomposition flow diagram;

FIG. 2 is a waveform diagram of a simulation signal;

FIG. 3 is an exploded view of the emulation signal EEMD;

FIG. 4 is an exploded view of the emulation signal MEEMD;

FIG. 5 is a flow chart of a gear fault feature vector construction;

FIG. 6 is a waveform of vibration signals for four operating states of the gearbox;

FIG. 7 is a flow chart of a whale optimization algorithm;

FIG. 8 is a flowchart of a WOA-LSSVM fault diagnosis model process;

FIG. 9 is a graph of four fault diagnosis model aliasing matrices;

FIG. 10 is a graph of fitness for three optimized models;

FIG. 11 is a flow chart schematic of a method of fault diagnosis of a planetary gearbox.

Detailed Description

The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.

Example 1

As shown in fig. 11, the present invention provides a fault diagnosis method of a planetary gearbox, comprising the steps of:

s1: acquiring a vibration signal of the planetary gearbox;

the MEEMD decomposition algorithm comprises the following steps:

s203: carrying out integration average on the jth order IMF component in the first processing signal and the jth order IMF component in the second processing signal to obtain a jth order integration average component;

s204: calculating the permutation entropy of the integrated average components;

s205: judging whether the permutation entropy calculated in the step S204 is greater than a preset threshold, if so, executing the step S206, otherwise, executing the step S207;

s207: and removing the abnormal signal from the vibration signal acquired in the step S201 to obtain a residual signal, and performing EMD decomposition on the residual signal to acquire an IMF function.

The amplitude of the white noise signal is within a range of 0.1 to 0.2 times the standard deviation of the vibration signal.

The computational expression for the integrated average component of the j-th order is:

where I (t) is the integrated average component of the j-th order, N is the length of the vibration signal, Ne is the logarithm of the white noise signal,

the IMF component of the j-th order in the first processed signal,

the IMF component of the j-th order in the second processed signal.

S3: extracting fault characteristics of the planetary gearbox according to the IMF function;

s4: and according to fault characteristics, carrying out classification and identification on fault types by adopting a pre-established and trained WOA-LSSVM classification model, wherein the WOA-LSSVM classification model is an LSSVM classification model for optimizing regularization parameters and kernel function parameters by adopting a whale optimization algorithm. The fault types identified by classification include normal, broken tooth, wear and pitting.

The training process of the WOA-LSSVM classification model comprises the following steps:

s401: determining a fitness function, wherein the fitness function is the classification accuracy of the training samples, and acquiring the training samples of the vibration signals of the planetary gear box;

s402: initializing parameters of a whale optimization algorithm and an LSSVM classification model;

s403: loading the training samples into an LSSVM classification model for model training, and calculating a fitness function of the LSSVM classification model;

s404: updating regularization parameters and kernel function parameters of the LSSVM classification model by adopting a whale optimization algorithm based on the self-adaptive function;

s406: and establishing a WOA-LSSVM classification model by adopting an optimal fitness function calculation value in an iteration process and corresponding regularization parameters and kernel function parameters.

The steps of the fault diagnosis method of the present embodiment are described in detail below.

1. Planetary gearbox fault extraction based on MEEMD algorithm and sample entropy

Empirical Mode Decomposition (EMD) suffers from modal aliasing. EEMD has a certain suppression effect on mode aliasing by adding white noise to the original signal. However, due to the limitation of the number of iterations of the EEMD algorithm, the decomposed components do not necessarily satisfy the IMF definition (there is a spurious component), and the physical significance of the instantaneous frequency of the component cannot be ensured. Different from EEMD algorithm, MEEMD algorithm adds paired white noise signals with opposite signs, reduces interference of the white noise signals, improves completeness of EEMD algorithm, and introduces permutation entropy on the basis. The MEEMD algorithm not only inhibits the modal aliasing phenomenon, but also limits the defects of the EEMD algorithm by calculating the permutation entropy of IMF components, separating abnormal signals of time sequence signals according to the permutation entropy value and a certain rule, and then carrying out EMD decomposition on the residual signals. Taking the original signal with length N as an example, the decomposition process is shown in fig. 1.

1.1) adding a pair of white noise signals with zero mean value into the original vibration signals x (t) of the gearbox respectively, wherein the amplitude of the white noise is 0.1 to 0.2 times of the standard deviation of the original signals mostly, and generating two x⁺(t) and x^-(t) two signals.

1.2) to x⁺(t) and x^-(t) EMD decomposition to obtain a first-order IMF component,

and

the component obtained by integration and averaging is

Wherein: ne is the logarithm of the white noise added; n is the length of the vibration signal, Ne in this document 50,.

1.3) calculating the signal I₁And (t) judging whether the entropy value is larger than a set threshold (0.6 is taken in the text) or not, and identifying the signal larger than the threshold as an abnormal signal, otherwise, identifying the signal as a normal signal.

1.4) if I₁(t) is an abnormal signal. Continue to execute steps 1.1), 1.2) and 1.3) until IMF component I_q(t) is a normal signal.

1.5) removing q-1 abnormal signals from the original signals of the gear box, performing EMD on the rest signals, and sequencing the obtained IMF components from high frequency to low frequency to complete the MEEMD of the signals.

Since the space limit permutation entropy calculation method is not shown here, in order to verify the superiority of the MEEMD algorithm compared with the EEMD algorithm, the simulation signal is set as formula (2), and the waveform is shown in fig. 2.

y(t)＝2sin(160πx)+3sin(20πx) (2)

The simulation signals are decomposed by two algorithms, EEMD and MEEMD, respectively, and the decomposition results are shown in FIG. 3 and FIG. 4. Table 1 shows indices of decomposition errors, orthogonality with respect to the original signal, and operation time in the two decomposition methods. From table 1, it is understood that the MEEMD decomposition effect is more excellent than the EEMD.

TABLE 1 EEMD comparison with MEEMD decomposition results

And applying the MEEMD algorithm with a better decomposition result to the decomposition of the vibration signal of the gearbox, selecting a proper IMF component, and constructing a fault characteristic vector of the gearbox, wherein the process is shown in FIG. 5. Taking the set of vibration signals of fig. 6 as an example, the fault feature vector is shown in table 2.

TABLE 2 MEEMD-sample entropy construction of Fault feature vectors

2. Planetary gearbox fault diagnosis model based on WOA-LSSVM

Compared with the traditional support vector machine, the LSSVM classification algorithm has the advantages that the objective function is changed from inequality constraint to equality constraint, and the complexity of calculation is greatly simplified. Experiments prove that the LSSVM algorithm obtains a good compromise between the complexity of the model and the accuracy of a training result, and is more suitable for the situation that a data sample is complex. The specific implementation method of the LSSVM is not described in detail herein.

According to the basic principle of the LSSVM algorithm, the fact that the acquisition of proper parameters is important for the LSSVM model can be known. LSSVMs have two parameters: the regularization parameter gamma and the kernel function parameter sigma greatly influence the LSSVM classification model. Some classical intelligent optimization algorithms are often used to optimize the LSSVM model, such as particle swarm optimization algorithm, genetic algorithm, etc. Although the algorithm has a certain improvement on the classification effect of the LSSVM, traps with local extreme values are not easy to jump out, so that the classification precision is not high, compared with the algorithms, the whale optimization algorithm has the characteristics of simple and quick operation, strong global search capability and the like, and has a great probability of getting rid of the local extreme values.

The implementation flow of the whale optimization algorithm is shown in fig. 7.

2.1) surrounding the prey

When whales search for a prey, the position of the prey should be determined firstly, then enclosure is carried out, and individuals in a group move to the optimal position on the assumption that the current optimal position is the target prey.

For the distance between the individual and the optimal whale position, the position is updated as follows:

where t represents the iteration of the current generation,

indicating the position of the best whale in the tth generation,

indicating the location of individual whales in the t-th generation. Wherein

And

is defined as follows:

in the formula

Represents [0,1]]T is the current iteration number, T_maxIs the maximum number of iterationsNumber when

By the time, the whale thinks that the prey is found, the air bubble attack can be started.

2.2) bubble attack

In the whale optimization algorithm, two whale predation modes are set, which are respectively as follows: a contraction enclosure mode and a spiral bubble net attack mode.

A contraction and enclosure manner: by reducing that in equation (4)

To achieve, known from the formula

Size of [ -a, a]In the meantime.

Spiral bubble net attack mode: the individual whales prey on the prey in a spiral path. The position equation is updated as follows.

In the formula

Representing the distance between whale and prey, b is a curve defining the form of a logarithmic spiral, l is [ -1,1 [ ]]A random number in between.

In order to simulate the mode of simultaneously using the contraction enclosure and the spiral path when a whale colony attacks a prey, a probability p is set by a whale optimization algorithm to set two predation modes of the contraction enclosure and the spiral path attack. p is a random number between [0,1], and assuming that the probability that whales respectively adopt two predation modes is 0.5, the whale positions are iterated by a mathematical model:

2.3) search for prey

In addition to the updated positions of the whales following the optimal positions, the whales can randomly update the positions of the whales during the predation process, so that the whales are forced to obtain a larger search range, and the Whale Optimization (WOA) algorithm has better global search capability. When in use

The whale randomly searches for prey behavior, and the mathematical expression at this stage is:

and (3) applying a Whale Optimization Algorithm (WOA) to the LSSVM algorithm to optimize and establish a fault diagnosis model of the planetary gearbox WOA-LSSVM. The implementation flow is as shown in fig. 8.

2.3.1) dividing the fault characteristic matrix of the planetary gearbox into a training sample matrix and a testing sample matrix according to a certain proportion.

2.3.2) determining a fitness function of the whale optimization algorithm, wherein the fitness function of the model is the classification accuracy of the planetary gearbox training samples, and finally, the fitness function is required to obtain the maximum value in all whale populations, namely the maximum classification accuracy rate, according to the finally selected parameters (the optimal positions of the whales).

2.3.3) initializing the parameters of the whale optimization algorithm and the least square support vector machine.

2.3.4) inputting two parameters of the LSSVM, and training the LSSVM model by using the training samples to calculate the fitness function of the LSSVM model.

2.3.5) Whale Optimization Algorithm (WOA) updates the optimal whale optimal position (parameter) and inputs into LSSVM model.

2.3.6) train a new LSSVM model using the training samples and calculate a fitness function.

2.3.7) loop through steps 2.3.4) to 2.3.6) until the maximum number of iterations for WOA is reached

2.3.8) records the parameters (whale position) corresponding to the optimal fitness function

2.3.9) using the optimal parameters to build a WOA-LSSVM planetary gearbox fault diagnosis model.

2.3.10) inputting the test sample into a WOA-LSSVM fault diagnosis model to complete the fault diagnosis of the planetary gearbox.

3. Experimental verification

The invention carries out tests according to a fault diagnosis platform of Shanghai electric group, sets the sampling frequency to be 2000 x 2.56hz, and adopts 4 vibration signals of normal, broken teeth, abrasion and pitting corrosion under the conditions of rotating speed of 880r/min and loading current of 0.05A through an acceleration sensor. Preprocessing signals to obtain the planetary gearbox in four states: normal, broken teeth, wear, pitting, vibration signals are 25 groups each, and 100 groups are total. In order to verify the superiority of the fault diagnosis model of the WOA-LSSVM planetary gearbox, the simulation contrast experiment process is set as follows:

3.1) randomly selecting gear box fault samples under four states: normal, broken tooth, abrasion and pitting corrosion are respectively 5 groups, and a total of 20 groups are used as training sample sets. The remaining 80 sets of samples were used as test samples, with 20 sets for each of the four failure types.

3.2) based on the theory and the method, respectively establishing four untrained fault diagnosis models of LSSVM, GA-LSSVM, PSO-LSSVM and WOA-LSSVM.

And 3.3) respectively inputting 20 groups of training samples into four fault diagnosis models, and training the models. And obtaining the trained model.

And 3.4) respectively inputting 80 groups of test samples into four planetary gearbox fault diagnosis models of LSSVM, GA-LSSVM, PSO-LSSVM and WOA-LSSVM which are trained, obtaining specific classification results, and then counting respective fault recognition rates.

The aliasing matrix of the fault diagnosis of the four planetary gearboxes of LSSVM, GA-LSSVM, PSO-LSSVM and WOA-LSSVM is shown in FIG. 9. The fitness function curves of the three optimization algorithms are shown in fig. 10.

From the aliasing matrix of fig. 9, the fault diagnosis model of LSSVM performs the worst, the classification of the WOA-LSSVM algorithm performs the best, and all four faults are correct. Analyzing the aliasing matrix to know that the fault types classified by mistake are wear and corrosion faults, and the two fault types are similar and are difficult to distinguish according to the practical situation, wherein the two faults are between: the total number of the fault diagnosis models of the LSSVM fault diagnosis model is 6, and the diagnosis models of the GA-LSSVM and the PSO-LSSVM fault diagnosis models are trapped in local extreme values, so that the number of the fault diagnosis models of the GA-LSSVM and the PSO-LSSVM fault diagnosis models is 5 and 4 respectively, and a good classification effect cannot be achieved.

Analyzing fig. 10, compared with two classic optimization algorithms, the WOA-LSSVM algorithm jumps out of local extremum and achieves better effect. As known from a WOA-LSSVM optimization model fitness curve, along with the increase of iteration times, the global search capacity of the whale optimization algorithm gradually appears, and local extreme values are continuously jumped out until the fitness obtains the maximum value. Compared with the WOA-LSSVM algorithm, the membership curve of the PSO-LSSVM and the genetic algorithm GA) is a straight line, and local extreme values cannot be jumped out.

The embodiment also provides a fault diagnosis device of the planetary gearbox, which comprises a memory and a processor, wherein the memory stores a computer program, and the processor calls the computer program to execute the steps of the method.

The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.

Claims

1. A method of fault diagnosis for a planetary gearbox comprising the steps of:

s1: acquiring a vibration signal of the planetary gearbox;

2. A method of fault diagnosis of an epicyclic gearbox according to claim 1, wherein said MEEMD decomposition algorithm comprises the following steps:

s201: adding a group of white noise signals with opposite amplitudes and zero mean value into the vibration signals respectively to obtain a first processing signal and a second processing signal respectively;

s204: calculating the permutation entropy of the integrated average components;

3. The method of diagnosing a malfunction of an epicyclic gearbox according to claim 2, wherein said white noise signal has an amplitude within the range of 0.1 to 0.2 times the standard deviation of said vibration signal.

4. A method as claimed in claim 2, wherein the calculation expression of the integrated average component of the j-th order is:

the IMF component of the j-th order in the first processed signal,

the IMF component of the j-th order in the second processed signal.

5. The method for diagnosing faults of an epicyclic gearbox according to claim 1, wherein in step S4, a pre-established and trained LSSVM classification model is used for classification and identification of fault types.

6. The method as claimed in claim 1, wherein in step S4, a WOA-LSSVM classification model is used to perform classification and identification of fault types, wherein the WOA-LSSVM classification model is an LSSVM classification model that uses whale optimization algorithm to optimize regularization parameters and kernel function parameters.

7. The method of claim 6, wherein the training process of the WOA-LSSVM classification model comprises the steps of:

8. The method as claimed in claim 7, wherein the fitness function is a classification accuracy of the training samples.

9. The method for diagnosing faults of an epicyclic gearbox according to claim 1, wherein said fault types identified by said classification in step S4 include normal, broken teeth, wear and pitting.

10. A fault diagnosis device for an epicyclic gearbox comprising a memory and a processor, said memory storing a computer program, said processor invoking said computer program for performing the steps of the method according to any of claims 1 to 9.