CN104614166B - Method for identifying failure state of rotor vibration signal of aircraft engine - Google Patents

Abstract

The present invention proposes a method for identifying the fault state of an aircraft engine rotor vibration signal, which mainly includes the following steps: first extracting 11 kinds of time domain, frequency domain and time-frequency domain features of the rotor vibration signal; then selecting appropriate features through clustering; Next, the classifier is trained using the clustering results; finally, new data is substituted into the trained classifier to generate classification results. The invention solves the problem that many irrelevant or redundant information exist in a large number of extracted diagnosis features due to subjective and objective reasons when identifying the fault state of the existing aircraft engine rotor. The present invention adopts the method of aircraft engine rotor vibration signal fault classification combined with clustering and classifier, and proposes a method framework for identification of aircraft engine rotor vibration signal fault state, which can automatically identify key features sensitive to fault features, and eliminate Correlation features that are not sensitive to fault features can effectively improve the fault state identification performance of aircraft engine rotor vibration signals.

Description

A method for fault state identification of aircraft engine rotor vibration signal

technical field

The invention belongs to the technical field of fault diagnosis and state monitoring of electromechanical systems, and specifically relates to a method for identifying fault states of aircraft engine rotor vibration signals.

Background technique

The aircraft operates in a complex environment and has high requirements for system reliability. The status monitoring, diagnosis and prediction of the aircraft system are important means to ensure the safety of the aircraft. For example, the aircraft system needs to monitor engine performance (such as patent: CN103370667A), landing gear status (such as patent: CN 103963986), elevator control (such as patent: CN 202987498), and whether various electronic and electromechanical equipment (such as patent: CN 102700718A) The status of each system such as normal operation. The overall state of these systems is affected by the sub-systems, for example, the state monitoring of anti-skid brake control box (such as patent: CN 104049630A), aircraft cables (such as patent: 103558513A) and engine thrust reverser (such as patent: CN 103979114A). These status information are sent to the aircraft's ACMS (Aircraft Condition Monitoring System, ACMS) system to monitor the status of each system and subsystem, and take corresponding alarm and isolation measures in case of failure. Condition monitoring and fault diagnosis of aircraft systems have become necessary means for flight safety and condition-based maintenance.

Aeroengine is the "heart" of modern aircraft. It works in the harsh environment of high pressure, high temperature and high load. The working condition of the engine directly affects the safety and reliability of the aircraft. The design of modern aeroengines pursues high thrust and high power-to-mass ratio, which reduces the size of the engine and increases the speed; at the same time, in order to ensure the efficiency of the compressor, the gap between the rotor and the stator is almost as small as the extreme. During the working process, once the whirl of the rotor is greater than the gap between the rotor and the stator, there will be friction between the rotor blades and the stator shell, which will affect the working stability of the rotor in a short period of time, and cause damage to the engine in severe cases.

Therefore, in the ground test of the engine, the vibration signal monitoring is often used as a means to observe its working state. When the rubbing condition is severe or obvious, the fault state can be easily judged no matter from the amplitude of the response signal or the appearance of the part. However, when slight friction occurs, it is difficult to judge from the sound, and it is not easy to identify by observing the appearance of the parts. Although there is currently a method to extract the features of the grinding fault and identify the fault state, due to subjective and objective reasons, there are many irrelevant or redundant information in the extracted diagnostic features, which leads to the gradual increase of the internal features of the data, which is manifested as The complexity of the distribution space of fault data state parameters, the repetition redundancy of data samples and the high degree of redundancy in the dimension of state parameters. High-dimensional data will generate large computational consumption during the processing process. At the same time, it is difficult to intuitively analyze the impact of some main attributes on the system for the fault problems determined by these multi-attributes, so as to find the essential changes and laws of faults. And the visual analysis of the structure, association and distribution of these data on the computer has brought great difficulties.

To sum up, some key features of aircraft engine rotor vibration signals are sensitive to faults and are independent of each other. They can provide complementary information and improve diagnostic accuracy, and should be fully utilized; some redundant or irrelevant features are not sensitive to faults and have no Utilizing value will increase the workload and cost of diagnosis and should be removed from it. Therefore, what is needed now is a method that can automatically select the key features of the aircraft engine rotor vibration signal, observe its working state, and identify the fault state.

Contents of the invention

In order to solve the problem of selecting key features in the fault characteristics in the prior aircraft engine rotor vibration signal fault state identification, the present invention proposes a method for aircraft engine rotor vibration signal fault state identification. Technical scheme of the present invention is:

The method for identifying a fault state of a rotor vibration signal of an aircraft engine is characterized in that it comprises the following steps:

Step 1: Collect the original vibration signal of the aircraft engine rotor with the fault label;

Step 2: Extract 11 features from the collected original vibration signal with fault label: mean value, variance, kurtosis, peak index, waveform index, pulse index, margin index, frequency peak ratio, frequency domain root mean square , the energy of the first and second nodes in the third layer after wavelet packet decomposition, the signal matrix is obtained

Wherein, the signal matrix dimension variable is d, d=1,2,...,D, D=11 represents the dimension size, t=1,2,...,T, T represents the number of signals;

Step 3: Transpose the signal matrix to transform the signal matrix into a characteristic matrix X _c : X _c =X ^T ;

Step 4: Use features as clustered data points for feature extraction:

Step 4.1: Construct the Mahalanobis distance matrix S _M :

in

X ¹ , X ² ,...,X ¹¹ are the column vectors of each column in the characteristic matrix X _c , μ ₁ , μ ₂ ,...,μ ₁₁ are the mean value of each column in the characteristic matrix X _c , 1 ^T is T×1 matrix and each element in the matrix is 1, E is the expectation, and median means to take the median;

Step 4.2: Set the maximum number of cycles N;

Step 4.3: Initialize a(i,j)=0, update r(i,j) using the following formula:

Among them, a(i,j) represents the cumulative evidence that i chooses j as its central point, r(i,j) represents the cumulative evidence that j is the central point of i, and a _p is the value of the last iteration of a(i,j) , is the value of the current iteration of a _p , and a _p+1 is The updated value of the current iteration of the corresponding r _p is the value of the last iteration of r(i,j), is the value of the current iteration of r _p , r _p+1 is The updated value of the current iteration of , λ is the convergence coefficient;

Step 4.4: Update a(i,j) using the following formula

Step 4.5: Determine the cluster center of feature i, and determine each feature according to j=arg max _j {a(i,j)+r(i,j)}(i=1,2,...,D) The attribution class and the center of each category: After the iteration is stopped, for feature i, if i≠j, it indicates that the cluster center of i is j; if i=j, it indicates that i itself is the cluster center;

Step 4.6: If the clustering center does not change for several consecutive iterations, or the number of iterations reaches the specified number N, then go to step 5, otherwise go to step 4.3;

Step 5: Use the clustering results with fault labels to train in the classifier to obtain the trained classifier;

Step 6: Extract the newly collected data with unknown results, perform feature extraction according to steps 2 to 4, and then use the classifier trained in step 5 to classify to generate classification results.

Beneficial effect

The aircraft engine rotor vibration signal fault state identification method adopted in the present invention can automatically identify key features sensitive to fault features, eliminate related features that are not sensitive to fault features, and do not rely on random selection of initial clustering points, instead all features They are all used as candidate representative points, and the clustering results are used as the input of the classifier for fault diagnosis.

The aircraft engine rotor vibration signal fault state identification method described in the present invention can find features sensitive to faults: on the one hand, it can use fewer features to identify faults, so as to reduce diagnostic costs and improve diagnostic efficiency; on the other hand, for As far as fault identification is concerned, it can shorten the training time, improve the generalization ability, or reduce the computational workload, and avoid the "curse of dimensionality" problem, thus providing a good method framework for the fault state identification of aircraft engine rotor vibration signals.

Description of drawings

Fig. 1 is the specific flowchart of the method of the present invention.

detailed description

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

General faults of aircraft engine rotors include: inner ring faults, outer ring faults, rolling element faults, unbalance faults. The method for identifying the fault state of the vibration signal of the aircraft engine rotor can identify these fault states, and the general process can be divided into two processes: offline and online: the offline process, that is, the engine rotor vibration data of the known fault category, through the method mentioned in the present invention Processing to obtain a trained classifier; online process, that is, to use the acceleration sensor on the aircraft to collect new vibration data from the engine rotor, and then use the method of vibration signal fault analysis to process the data and substitute it into the trained classifier, and finally Displays fault status results.

The method for identifying the fault state of the aircraft engine rotor vibration signal described in this embodiment, the specific process is as shown in Figure 1:

Step 1: Use the acceleration sensor installed on the rotor to collect the original vibration signal of the aircraft engine rotor with the fault label.

Step 2: Extract 11 features from the collected original vibration signal with fault label: mean value, variance, kurtosis, peak index, waveform index, pulse index, margin index, frequency peak ratio, frequency domain root mean square , the energy of the first and second nodes in the third layer after wavelet packet decomposition, the signal matrix is obtained

Wherein, the dimension variable of the signal matrix is d, where d=1, 2, .

Step 3: Transpose the signal matrix to transform the signal matrix into a characteristic matrix X _c : X _c =X ^T .

Step 4: Use features as clustered data points for feature extraction:

This clustering method does not require to specify the number of clusters and the initialization center, but only needs to input the Mahalanobis distance matrix S between data points. s(i,j) indicates the possibility of i being classified into the class centered on j, the larger the value, the more likely i is classified into this class. When i=j, s(j,j) indicates the possibility of j being the center of the class. Obviously, the larger s(j,j) is, the more likely j is to be the center. Usually according to a certain measure, after calculating all s(i, j) (i≠j), s(j, j) can take a constant, if all the data points have the same possibility as the center point at the beginning, then take The same constant, it should be noted that the value of s(j,j) directly affects the clustering results, in general, the larger s(j,j), the more clusters will be generated in the end.

Step 4.1: Use the covariance matrix formula and the Mahalanobis distance matrix formula to calculate the Mahalanobis distance between the column vectors of the feature matrix and the column vectors, construct the Mahalanobis distance matrix S _M , and let each diagonal position of S _M take a value is the median of the corresponding row of the _SM matrix:

The Mahalanobis distance matrix formula is

The covariance matrix formula is

X ¹ , X ² ,...,X ¹¹ are the column vectors of each column in the characteristic matrix X _c , μ ₁ , μ ₂ ,...,μ ₁₁ are the mean value of each column in the characteristic matrix X _c , 1 ^T is T×1 matrix and each element in the matrix is 1, E is the expectation, and median means to take the median.

Step 4.2: Set the maximum number of cycles N=50.

Step 4.3: Initialize a(i,j)=0, update r(i,j) using the following formula:

Among them, a(i,j) represents the cumulative evidence that i chooses j as its central point, r(i,j) represents the cumulative evidence that j is the central point of i, and a _p is the value of the last iteration of a(i,j) , is the value of the current iteration of a _p , and a _p+1 is The updated value of the current iteration of the corresponding r _p is the value of the last iteration of r(i,j), is the value of the current iteration of r _p , r _p+1 is The updated value of the current iteration of , λ is the convergence coefficient, which is used to adjust the convergence speed of the algorithm and the stability of the iterative process, the larger the λ, the better the effect of eliminating oscillation, but the slower the convergence speed, and vice versa, where λ= 0.5.

Step 4.4: Update a(i,j) using the following formula

Step 4.5: Determine the cluster center of feature i, and determine each feature according to j=arg max _j {a(i,j)+r(i,j)}(i=1,2,...,D) The belonging category and the center of each category: After the iteration is stopped, for feature i, if i≠j, it indicates that the cluster center of i is j; if i=j, it indicates that i itself is the cluster center.

Step 4.6: If the clustering result is stable, that is, the clustering center does not change for several consecutive iterations, or the number of iterations reaches the specified number N, then proceed to step 5, otherwise perform step 4.3, and finally several clustering centers can be automatically generated, and at the same time Classify the remaining data points into the corresponding classes. The feature corresponding to the cluster center is selected as the result of feature selection, so as to find out which features are the centers of all features and eliminate redundant features.

Step 5: Use the clustering results with fault labels to train in the classifier to obtain a trained classifier.

Step 6: Extract the newly collected data with unknown results, perform feature extraction according to steps 2 to 4, and then use the classifier trained in step 5 to classify to generate classification results.

The analysis and implementation ideas of the method of the present invention are obviously not limited to the problem of identification of the failure state of the aircraft engine rotor vibration signal. Adaptive adjustments can also be made for the identification of signal fault states in other complex systems, so as to carry out research on fault state identification methods based on system state monitoring data, and provide more effective decision-making support methods for comprehensive support and maintenance of various complex systems.

Claims (1)

1. a method for aircraft engine rotor vibration signal fault state recognition, is characterized in that: comprise the following steps: Step 1: Collect the original vibration signal of the aircraft engine rotor with the fault label; Step 2: Extract 11 features from the collected original vibration signal with fault label: mean value, variance, kurtosis, peak index, waveform index, pulse index, margin index, frequency peak ratio, frequency domain root mean square , the energy of the first and second nodes in the third layer after wavelet packet decomposition, the signal matrix is obtained Wherein, the signal matrix dimension variable is d, d=1,2,...,D, D=11 represents the dimension size, t=1,2,...,T, T represents the number of signals; Step 3: Transpose the signal matrix to transform the signal matrix into a characteristic matrix X _c : X _c =X ^T ; Step 4: Use features as clustered data points for feature extraction: Step 4.1: Construct the Mahalanobis distance matrix S _M : in $\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&NotEqual;</span>\mathrm{<span\; class="notranslate">\; j</span>}\\ {\mathrm{<span\; class="notranslate">\; median</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&NotEqual;</span>\mathrm{<span\; class="notranslate">\; j</span>}}\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; j</span>}\end{array}\right.\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 11</span>}$ X ¹ , X ² ,...,X ¹¹ are the column vectors of each column in the characteristic matrix X _c , μ ₁ , μ ₂ ,...,μ ₁₁ are the mean value of each column in the characteristic matrix X _c , 1 ^T is T×1 matrix and each element in the matrix is 1, E is the expectation, and median means to take the median; Step 4.2: Set the maximum number of cycles N; Step 4.3: Initialize a(i,j)=0, update r(i,j) using the following formula: $\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; \&NotEqual;</span>\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>$ $\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; old</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}\\ {\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; old</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}\end{array}\right.$ Among them, a(i,j) represents the cumulative evidence that i chooses j as its central point, r(i,j) represents the cumulative evidence that j is the central point of i, and a _p is the value of the last iteration of a(i,j) , is the value of the current iteration of a _p , and a _p+1 is The updated value of the current iteration of the corresponding r _p is the value of the last iteration of r(i,j), is the value of the current iteration of r _p , r _p+1 is The updated value of the current iteration of , λ is the convergence coefficient; Step 4.4: Update a(i,j) using the following formula $\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; \&NotElement;</span><span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \}</span>}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&NotEqual;</span>\mathrm{<span\; class="notranslate">\; j</span>}\\ \underset{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; \&NotEqual;</span>\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; j</span>}\end{array}\right.<span\; class="notranslate">\; ;</span>$ Step 4.5: Determine the cluster center of feature _i , and determine each feature’s Belonging category and the center of each category: After the iteration is stopped, for feature i, if i≠j, it indicates that the cluster center of i is j; if i=j, it indicates that i itself is the cluster center; Step 4.6: If the clustering center does not change for several consecutive iterations, or the number of iterations reaches the specified number N, then go to step 5, otherwise go to step 4.3; Step 5: Use the clustering results with fault labels to train in the classifier to obtain a trained classifier; Step 6: Extract the newly collected data with unknown results, perform feature extraction according to steps 2 to 4, and then use the classifier trained in step 5 to classify to generate classification results.