CN100485342C - Integrated supporting vector machine mixed intelligent diagnosing method for mechanical fault - Google Patents

Abstract

The present invention discloses an integrated support vector machine mixed intelligent diagnosis method of machine failure. Said method includes the following steps: respectively adopting and lifting small wave packet, utilizing frequency band and empirical mode decomposition process to decompose vibration signal according to the eigen mode component and extract time domain statistical character of decomposed signal to form total characteristic set; providing characteristic distance evaluation technique and characteristic evaluation index; utilizing said characteristic evaluation index to select most sensitive characteristic set from the total characteristic set; using said most sensitive characteristic set as diagnosis characteristics and creating integrated support vector machine mixed intelligent diagnosis model so as to implement intelligent diagnosis of machine failure state.

Description

The integrated supporting vector machine mixed intelligent diagnosing method of mechanical fault

Technical field

The invention belongs to mechanical equipment fault intelligent diagnostics field, be specifically related to a kind of integrated supporting vector machine mixed intelligent diagnosing method of mechanical fault.

Background technology

In order to break away from the problem that mechanical fault diagnosis is too dependent on professional and technical personnel and diagnostician, efficient, the reliable intelligent diagnostics of canbe used on line, in recent years, people are applied to the intelligent trouble diagnosis of plant equipment with artificial intelligence technologys such as fuzzy theory, expert system, neural network and cluster analyses, have obtained certain effect in practice.But it is found that in actual applications these technology no all roses, for example: the fuzzy fault diagnosis often need manually be determined subordinate function and fuzzy relation matrix by priori, but in fact, there are many difficulties in subordinate function and the fuzzy relation matrix that obtains to conform to the equipment actual conditions; Also be faced with many difficulties in the expert system, as lack the expression of effective fault diagnosis knowledge, experimental knowledge is obtained difficulty etc., and in addition, the operating personnel of expert system also need to possess higher level, is difficult for popularizing; The master sample of some need be provided during neural network classification, the master sample that obtains all faults of equipment is quite to be difficult for, in addition, neural network all needs raw data is carried out feature extraction in actual applications, if do not do feature extraction, and directly participate in calculating with raw data, the structure of network will be very huge, can't satisfy the requirement of real time on-line monitoring diagnosis; Cluster analysis can not need the failure criterion sample, it forms a plurality of classes with data sample, data sample difference between the requirement class should be big as much as possible, and the difference between the data sample should be as much as possible little in the same class, promptly satisfy the principle of " similarity between minimized class; similarity in the maximized class ", therefore, undesirable to the cluster effect of early stage Weak fault or combined failure.

In engineering practice, fault diagnosis for large complicated plant equipment, often measurand is many, the data volume of analyzing and processing is big, carry out the fault sample famine of intelligent diagnostics model training, and plant equipment is the system of a complexity, time variation, randomness, many-sided factor such as ambiguity makes fuzzy theory, expert system, traditional intelligent diagnosing method such as neural network and cluster analysis is difficult to the malfunction of equipment operation is made identification accurately and effectively, particularly for potential, early stage Weak fault lacks effective recognition and diagnostic means.In order to solve the existing problem in intelligent trouble diagnosis field, urgently need research and introduce new theory and technology, new, intelligent diagnosis technology and method are efficiently proposed.

Support vector machine is a kind of novel intelligence learning machine that development in recent years is got up, and is suitable for the mechanical disorder pattern-recognition of few sample, and has obtained certain achievement in fault diagnosis.Because the high complexity of space-time, the execution of support vector machine usually is a kind of approximate treatment yet in actual applications; Simultaneously, the selection problem of kernel function parameter makes support vector machine that study very easily take place or owes to learn phenomenon, directly influences it and promotes performance.

Integrated supporting vector machine can improve the classification performance of single support vector machine effectively, yet, when very faint or fault sample is not true to type when the feature of initial failure, still may produce wrong diagnostic result.So need improve the quality of diagnostic message by the modern signal processing method in the link of feature extraction, thereby improve the accuracy and the reliability of fault diagnosis.Select suitable modern signal analytical approach, the fault that machine essential fault, initial failure etc. are difficult to indicate all may be discerned, for the intelligent diagnostics of early stage incipient fault provides foundation.Integrated use artificial intelligence and systematic analytic method advantage is separately learnt from other's strong points to offset one's weaknesses, and has complementary advantages, can improve the rapidity and the accuracy of diagnosis effectively, reduce misdiagnosis rate and rate of missed diagnosis, determine correct time and position that fault takes place, estimate its order of severity and development trend.

Summary of the invention

The objective of the invention is to overcome above-mentioned the deficiencies in the prior art, a kind of integrated supporting vector machine mixed intelligent diagnosing method of mechanical fault is provided, this method precision height, cost are low, simple and reliable, be convenient to use in the engineering practice, by Lifting Wavelet packet transform and these two kinds of modern signal processing technology of empirical mode decomposition, characteristic distance assessment technology and integrated supporting vector machine are effectively merged, realize intelligent diagnostics to the failure operation state.

Technical scheme of the present invention is to solve like this: undertaken by following step:

1) adopts the Lifting Wavelet bag by the eigenmodes component vibration signal to be decomposed respectively, extract the time domain statistical nature of decomposed signal, constitute all feature sets by frequency range and empirical mode decomposition;

2) propose characteristic distance assessment technology and feature evaluation index, from all feature sets, choose sensitive features collection according to feature evaluation index size;

The sensitive features collection that 3) will obtain is set up the integrated supporting vector machine mixed intelligent diagnostic model as diagnostic characteristic, realizes the intelligent diagnostics to equipment running status.

Said characteristic distance assessment technology and feature evaluation index are as follows:

At first, for an original vibration signal X (t), calculate 1 time domain statistical nature F _Time, 1 time domain statistical nature is meant one or more in average, variance, root amplitude, effective value, peak value, measure of skewness, kurtosis, waveform index, peak value index, nargin index and the pulse index; Utilize the Lifting Wavelet packet transform that X (t) is decomposed the m layer, the m value is 2,3 or 4, obtains 2 ^mIndividual decomposed signal calculates 1 statistical nature respectively to each decomposed signal, obtains 1 * 2 altogether ^mFeature F behind the individual Lifting Wavelet packet transform _WpAdopt empirical mode decomposition that X (t) is decomposed and obtain preceding n eigenmodes component, the n value is 4 to 8 integer, calculates the relative energy feature F of n eigenmodes component _EmdThereby, constitute all feature set F _Total=[F _TimeF _WpF _Emd];

Then, with all feature set F _TotalThe class interval S _bWith in the class apart from S _wRatio be set at apart from evaluation index J _A

At last, according to feature evaluation index J _ASize, from all feature set F _TotalThe middle feature evaluation index J that selects greater than given threshold value ρ _APairing feature is made as ρ with the pairing threshold value of maximum classification accuracy, thereby constitutes sensitive features collection F _Sensitivity

Described integrated supporting vector machine mixed intelligent diagnostic model method for building up is as follows:

At first utilize packed algorithm, be called for short the Bagging algorithm, from training sample, generate T training sample subclass TR at random _t(t=1,2 ..., T), the T value is 10 to 100 integer, utilizes T multi-category support vector machines to these subclass TR then _tTrain, obtain T sub-classifier f _tIf T sub-classifier f _tAverage error in classification be

, will $f=1/\hat{E}$ Be the fitness function of genetic algorithm, to T sub-classifier f _tIntegrated result be optimized, obtain the weight vector w of the optimization weights of sorter, thereby constitute integrated supporting vector machine, with sensitive features collection F greater than predetermined threshold value λ=1/T _SensitivityAs diagnostic characteristic, integrated supporting vector machine is trained, can set up the hybrid intelligent diagnostic model, utilize the hybrid intelligent diagnostic model to realize the intelligent diagnostics of mechanical fault.

Because the present invention has realized the feature extraction based on the modern signal analytical technology, the selection and the mixing of integrated intelligent sorting technique on algorithm of sensitive features collection, the present invention has the following significant advantage that is different from classic method:

1) carries out characteristic distance assessment and can remove the uncorrelated or redundant information that primitive character is concentrated effectively, choose sensitive features collection, thereby improve the classification accuracy and the operation efficiency of sorter;

2) set up the integrated supporting vector machine mixed intelligent model and carry out intelligent classification, can carry out few sample mode identification well, improved the classification performance and the anti-noise ability of single sorter effectively, for diagnosis accurately and effectively provides assurance;

3) whole process has realized feature extraction, feature selecting and the pattern-recognition mixing on algorithm, for the intelligent diagnostics of mechanical fault provides effective practical new technology.

Description of drawings

Fig. 1 is the integrated supporting vector machine mixed intelligent diagnostic flow chart of mechanical fault;

Fig. 2 is that certain electric locomotive traveling portion wheel is to structural representation;

Fig. 3 be sensitive features collection select apart from evaluation graph;

Fig. 4 is under the different feature evaluation metrics-thresholds, the classification performance comparison diagram of integrated supporting vector machine and single support vector machine and Bagging algorithm;

Fig. 5 is under the different noise content, the classification performance comparison diagram of integrated supporting vector machine and single support vector machine and Bagging algorithm.

Embodiment

Accompanying drawing is specific embodiments of the invention;

Below in conjunction with accompanying drawing content of the present invention is described in further detail:

1) the vibratory output signal to equipment carries out Lifting Wavelet packet transform and empirical mode decomposition respectively, extract the time domain statistical nature of each different frequency range component, constitute all feature sets, all feature sets that obtain are carried out the characteristic distance assessment, choose sensitive features collection;

The sensitive features collection that 2) will obtain is trained integrated supporting vector machine as diagnostic characteristic, and the hybrid intelligent diagnostic model that trains can carry out intelligent classification to equipment running status.

According to the hybrid intelligent diagnostic structure flow process of foregoing invention content and Fig. 1, at first, structure primitive character collection.

For an original vibration signal X (t), extract 11 time domain statistical nature F _Time(average, variance, root amplitude, effective value, peak value, measure of skewness, kurtosis, waveform index, peak value index, nargin index and pulse index).

For original vibration signal X (t), data length is L, and the j yardstick is with approximation signal { s _J+1(k) } subdivision is strange, even sample sequence { s _j(2k+1) } and { s _j(2k) }, adopt the interpolation subdividing principle to predict strange sample, then detail signal d with N even sample _j(k) be

$d_j\left(k\right)=s_{j+1}(2k+1)-\underset{m=1}{\overset{N}{\mathrm{\Σ}}}p\left(m\right)s_{j+1}(2m+k-N)---\left(1\right)$

P (m) is a predictive coefficient in the formula (1), order: P=[p (1) ..., p (N)] ^T, employing formula (1) is tried to achieve

Individual detail signal d _j(k) upgrade even sample { s _j(2k) }, approximation signal s then _j(k) be

$s_j\left(k\right)=s_{j+1}\left(2k\right)-\underset{m=1}{\overset{\tilde{N}}{\mathrm{\Σ}}}u\left(m\right)d_j(m+k-\tilde{N}/2-1)---\left(2\right)$

U (m) is a update coefficients in the formula (2), order: $U={[u\left(1\right),\·\·\·,u\widetilde{\left(N\right)}]}^T$ 。In formula (1) and formula (2), choose different N and

The scaling function and the wavelet function of different vanishing moments will be obtained having.

With N and

The small echo with impact signal characteristic of structure is a basis function, adopts to promote strategy to wavelets Subspace W _jDecompose, thereby obtain subspace behind the wavelet package transforms

Signal be X _j={ x _{J, n, l}, j, n, 1 ∈ Z}, x _{J, n, l}The 1st data for the n frequency band of j yardstick.

To original signal X (t), utilize the Lifting Wavelet packet transform to decompose 3 layers, obtain 8 sections wavelet packet frequency band coefficients, to every section coefficient each extract with the original signal processing in 11 identical statistical natures, obtain 88 feature F behind the Lifting Wavelet packet transform altogether _Wp

Empirical mode decomposition can be decomposed into arbitrary signal several eigenmodes components and a remainder.So-called eigenmodes component is exactly function or the signal that satisfies 2 conditions: 1. in whole data sequence, the quantity of extreme point must equate with the quantity of crossing at 0 or differ one at most.2. in any point, the average of the lower envelope line that coenvelope line that the local maximum point of data sequence is determined and local minizing point determine is 0, and promptly signal is about the local symmetry of time shaft.The essence of empirical mode decomposition is the process of a screening, and through after a series of decomposition, time series X (t) can be expressed as n eigenmodes component f _i(t) and a remainder r _n(t) sum, promptly

$x\left(t\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{f}_i\left(t\right)+r_n\left(t\right)---\left(3\right)$

N the eigenmodes component f that obtains in the formula (3) _i(t), its frequency is arranged from big to small, f ₁(t) contained frequency is the highest, f _n(t) contained frequency is minimum, remainder r _n(t) be the monotonic sequence of a non-concussion.

To original signal X (t), utilize empirical mode decomposition to obtain the first six eigenmodes component, extract the relative energy feature F of 6 eigenmodes components _Emd, obtain all feature set F _Total=[F _TimeF _WpF _Emd].

Then, to primitive character collection F _TotalCarry out feature selecting.

Suppose c mode class ω ₁, ω ₂..., ω _cThe associating set of eigenvectors be { q ^{(i, k)}, i=1,2 ..., c; K=1,2 ..., N _i, q wherein ^{(i, k)}Be k feature among the ω i, N _iBe ω _iThe number of middle proper vector.Feature selecting can be divided into three steps:

The first step: calculate ω _iMean distance in the class between all proper vectors is as follows

$S_i=\frac{1}{2N_i}\underset{j=1}{\overset{N_j}{\mathrm{\Σ}}}\frac{1}{N_i-1}\underset{k=1}{\overset{N_i}{\mathrm{\Σ}}}|q^{(i,j)}-q^{(i,k)}|---\left(4\right)$

To S _i(i=1,2 ..., c) ask and obtain that distance is in the average class after average

$S_w=\frac{1}{c}\underset{i=1}{\overset{c}{\mathrm{\Σ}}}{S}_i---\left(5\right)$

Second step: the between class distance that calculates c mode class is as follows

$S_b=\frac{1}{c}\underset{i=1}{\overset{c}{\mathrm{\Σ}}}|{\mathrm{\μ}}^{\left(i\right)}-\mathrm{\μ}|---\left(6\right)$

Wherein: ${\mathrm{\μ}}^{\left(i\right)}=\frac{1}{N_i}\underset{k=1}{\overset{N_i}{\mathrm{\Σ}}}{q}^{(i,k)}$ Be ω _iIn the average of all features, $\mathrm{\μ}=\frac{1}{c}\underset{i=1}{\overset{c}{\mathrm{\Σ}}}\frac{1}{N_i}\underset{k=1}{\overset{N_i}{\mathrm{\Σ}}}{q}^{(i,k)}$ Population mean for c mode class sample.

The 3rd step: the ratio J of distance in definition class spacing and the class _ABe the distance evaluation index

$J_A=\frac{S_b}{S_w}---\left(7\right)$

As can be seen, distance and big average between class distance just have good separability in the little average class from the definition of (7) formula, therefore select the J greater than certain threshold value _APairing feature constitutes sensitive features collection F _Sensitivity, the pairing threshold value of maximum classification accuracy is made as ρ.

At last, set up the integrated supporting vector machine mixed intelligent diagnostic model, to sensitive features collection F _SensitivityCarry out intelligent classification.

(Support Vector Machines SVMs) is based on a kind of small sample algorithm for pattern recognition of Statistical Learning Theory and structural risk minimization to support vector machine.In the integrated supporting vector machine algorithm, at first utilize the Bagging algorithm from training sample, to generate T training sample subclass TR at random _t(t=1,2 ..., T), utilize T multi-category support vector machines then to these subclass TR _tTrain, obtain T sub-classifier f _tIn the present invention, establish T sub-classifier f _tAverage error in classification be

, will $f=1/\hat{E}$ As the fitness function of genetic algorithm, to T sub-classifier f _tIntegrated result be optimized, obtain the weight vector w of the optimization weights of sorter, thereby constitute integrated supporting vector machine greater than predetermined threshold value λ=1/T.Its algorithm is as shown in table 1.

Table 1 integrated supporting vector machine algorithm

The supporting vector machine mixed intelligent diagnostic model that utilization trains can be diagnosed equipment running status, thereby can obtain the type of fault.

With reference to shown in Figure 1, be the integrated supporting vector machine mixed intelligent diagnostic flow chart of mechanical fault, the vibration signal of plant equipment picks up through sensor and data acquisition system (DAS), and after the pre-service of signal pre-processing system, can obtain original time domain vibration signal; 88 statistical natures after 11 temporal signatures, Lifting Wavelet bag that calculate original time domain vibration signal decompose and 6 relative energy features after the empirical mode decomposition constitute all feature sets; Utilize the characteristic distance assessment technology that all feature sets are carried out feature selecting, constitute sensitive features collection; Sensitive features set pair integrated supporting vector machine with training sample is trained, and makes up the integrated supporting vector machine mixed intelligent diagnostic model, utilizes this model to carry out the real-time intelligent diagnosis to plant equipment, obtains the hybrid intelligent diagnostic result at last.

With reference to shown in Figure 2, for certain electric locomotive traveling portion wheel to structural representation, each is taken turns being made up of two bearings 1 and 5, two wheels 2 and 4 and axletrees 3, degree of will speed up sensor 6 is installed in and carries out vibration signal above the bearing seat and pick up.

With reference to shown in Figure 3, for sensitive features collection select apart from evaluation graph, according to preset threshold ρ, thus can determine greater than this threshold value apart from evaluation index J _APairing feature is sensitive features collection, and wherein, horizontal ordinate is the feature sequence number, and ordinate is apart from evaluation index J _A

With reference to shown in Figure 4, under different feature evaluation metrics-thresholds, integrated vector machine and the single classification performance comparison diagram of holding vector machine and Bagging algorithm held, by with classic method more as can be seen, under different threshold values, the present invention has good classifying quality, and the feature of selecting when ρ=20 is sensitive features collection, wherein, horizontal ordinate is threshold value ρ, and ordinate is a classification accuracy.

With reference to shown in Figure 5, under different noise content, the classification performance comparison diagram of integrated SVMs and single SVMs and Bagging algorithm, by with classic method more as can be seen, under different noise content, the present invention has very high diagnosis efficiency height and anti-noise ability, wherein, horizontal ordinate is a noise content number percent, and ordinate is a classification accuracy.

Embodiment:

This embodiment has provided the specific implementation process of the present invention in engineering practice, simultaneous verification should the invention validity.

Certain passenger-carrying version electric locomotive traveling portion by six accessory whorls to forming.Wheel is to being made up of an axletree and two wheels, and every accessory whorl is to combining with axle box again.Wheel to structure as shown in Figure 2.Carry out signals collecting by the acceleration transducer that is installed in axletree bearing top, sample frequency is 12.8KHz, and data length is 8192 points.

Get this electric locomotive bearing each 36 groups of four kinds of operating modes (normal, outer ring fault, rolling body fault, outer ring and rolling body combined failure) vibration datas down, wherein 22 groups as training data, other 14 groups as test sample book.Concentrate all feature set F that extract from the training data of electric locomotive bear vibration _Total=[F _TimeF _WpF _Emd] form by 105 features, these features apart from the evaluation index value as shown in Figure 3.

In order to verify the classification performance of integrated SVMs method, also analyze simultaneously to single SVMs sorting algorithm with based on the Bagging algorithm classification algorithm of SVMs.Get the number T=20 of integrated SVMs herein, can select the predetermined threshold value λ=1/T=0.05 among the integrated SVMs.

The classification results of single SVMs, Bagging algorithm and integrated SVMs is as shown in table 2.Wherein nicety of grading is that each test repeats 10 times average result.

SVMs, the Bagging of the different threshold value correspondences of table 2 and the nicety of grading of integrated Bagging are relatively

As shown in table 2, along with the increase of threshold value, number of features reduces gradually.In training process, the nicety of grading of single SVMs is all greater than 98.95%, and in ρ=0.5,5 or obtained maximal value (100%) at 20 o'clock; Yet the nicety of grading of the training sample of Bagging algorithm and integrated SVMs all is 100%.It can also be seen that in addition in integrated SVMs, the integrated number (9.3) of average SVMs is approximately half of overall number (20), therefore compares with the Bagging algorithm, can select the test duration of integrated SVMs to reduce significantly (dropping to 0.02 second from 0.12 second).

At different feature selecting threshold values, the test sample book classification results of single SVMs, Bagging algorithm and integrated SVMs as shown in Figure 4.

As can be seen from Figure 4, for the test result of single SVMs, when not carrying out feature selecting, its classification accuracy only is 63.33%; Classification accuracy improves along with the increase of threshold value, is up to 97.05%, this moment threshold value ρ=20, this feature that shows that ρ=20 o'clock are selected is sensitive features collection.Can find that the classification performance of integrated SVMs is better than other two kinds of methods, when ρ=20, integrated SVMs obtains best classifying quality (100%).

Further verify the extensive performance of integrated SVMs, in test data, add the random noise of different content.Be without loss of generality, the number T=20 that gets integrated SVMs in Bagging algorithm and integrated SVMs experimentizes, threshold value ρ=20.The classification accuracy of different noise content as shown in Figure 5.As can be seen from the figure, along with the increase of noise content, for integrated SVMs, when noise content less than 20% the time, its nicety of grading is all greater than 90%.When noise content was between 0 and 40%, the nicety of grading of integrated SVMs all was higher than the Bagging algorithm.Further increase along with noise content, the nicety of grading of the nicety of grading of integrated SVMs and Bagging algorithm is suitable, all be about 70%, but in integrated SVMs, the integrated number (9.3) of average SVMs is approximately half of integrated number (20) of Bagging algorithm, can select the test duration of integrated SVMs to reduce significantly (dropping to 0.02 second from 0.12 second).These illustrate that all integrated SVMs counting yield height, anti-noise ability are strong.

From the analysis of Fig. 4 and Fig. 5 as can be known, the classification performance of integrated SVMs is better than single SVMs and Bagging algorithm, fault diagnosis for the locomotive wheel set bearing, when the threshold setting in the characteristic distance assessment technology is ρ=20, its classification accuracy is 100%, this explanation integrated supporting vector machine mixed intelligent diagnosing method can be diagnosed out outer ring fault, rolling body fault and outer ring and the rolling body combined failure of bearing exactly, has good intelligent diagnostics ability.

Claims (1)

1. the integrated supporting vector machine mixed intelligent diagnosing method of a mechanical fault comprises:

1) adopts the Lifting Wavelet bag by the eigenmodes component original vibration signal to be decomposed respectively, extract the time domain statistical nature of decomposed signal and original vibration signal, constitute all feature sets by frequency range and empirical mode decomposition;

It is characterized in that:

2) propose ratio with distance in the class spacing of all feature sets and the class as the distance evaluation index, from all feature sets, choose sensitive features collection according to distance evaluation index size;

At first, for an original vibration signal X (t), calculate l time domain statistical nature F _Time, l time domain statistical nature is meant one or more in average, variance, root amplitude, effective value, peak value, measure of skewness, kurtosis, waveform index, peak value index, nargin index and the pulse index; Utilize the Lifting Wavelet packet transform that X (t) is decomposed the m layer, the m value is 2,3 or 4, obtains 2 ^mIndividual decomposed signal calculates l statistical nature respectively to each decomposed signal, obtains l * 2 altogether ^mFeature F behind the individual Lifting Wavelet packet transform _WpAdopt empirical mode decomposition that X (t) is decomposed and obtain preceding n eigenmodes component, the n value is 4-8, calculates the relative energy feature F of n eigenmodes component _EmdThereby, constitute all feature set F _Total=[F _TimeF _WpF _Emd];

Then, with all feature set F _TotalThe class interval S _bWith in the class apart from S _wRatio be set at apart from evaluation index J _A

At last, according to distance evaluation index J _ASize, from all feature set F _TotalIn select greater than given threshold value ρ apart from evaluation index J _APairing feature is made as ρ with the pairing threshold value of maximum classification accuracy, thereby constitutes sensitive features collection F _Sensitivity

The sensitive features collection that 3) will obtain is set up the integrated supporting vector machine mixed intelligent diagnostic model based on packed algorithm and genetic algorithm as diagnostic characteristic, realizes the intelligent diagnostics to equipment running status;

At first utilize packed algorithm from training sample, to generate T training sample subclass TR at random _t(t=1,2 ..., T), the T value is 10-100, utilizes T multi-category support vector machines to these subclass TR then _tTrain, obtain T sub-classifier f _tIf T sub-classifier f _tAverage error in classification be

Will $f=1/\hat{E}$ As the fitness function of genetic algorithm, to T sub-classifier f _tIntegrated result be optimized, obtain the weight vector w of the optimization weights of T sub-classifier, thereby constitute integrated supporting vector machine, with sensitive features collection F greater than predetermined threshold value λ=1/T _SensitivityAs diagnostic characteristic, integrated supporting vector machine is trained, can set up the hybrid intelligent diagnostic model, utilize the hybrid intelligent diagnostic model to realize the intelligent diagnostics of mechanical fault.

Images