Bearing fault diagnosis method based on width incremental learning

Abstract

The invention discloses a bearing fault diagnosis method based on width incremental learning. The method comprises the steps that bearing original time domain data needed by a width learning fault classification model is acquired; original time domain data of the bearing is preprocessed and then divided into a new data set and an old data set, and the old data set is input into a width learning fault classification model for training; meanwhile, SVDD anomaly detection model training is carried out on the old data set; secondly, testing whether the new data set is abnormal type data or not by using an SVDD (Singular Value Decomposition Decomposition) anomaly detection algorithm; and finally, performing width increment learning of sample increment or class increment according to an anomaly detection result to achieve an effect of updating the model. According to the method, initial anomaly detection is carried out on new data by utilizing SVDD; and meanwhile, according to an anomaly detection result, width stacking is performed on width learning to realize a sample increment or class increment effect of the model, and high real-time performance and high accuracy of the model to new data are met.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing detection, and in particular to a bearing fault diagnosis method based on width increment learning. Background Art

[0002] Due to the important role that bearings play in modern mechanical equipment, they are widely used in various rotating machinery, such as fans, generators, machine tools and other equipment. Once a failure occurs, it will seriously affect the normal operation of the rotating machinery, which is closely related to the safety and reliability of the mechanical equipment. Therefore, studying how to diagnose bearing faults and timely diagnose bearing fault types has become a hot topic in research.

[0003] In order to adapt to the intelligent development of mechanical equipment and the characteristics of big data, and effectively tap the huge potential value of monitoring data, data-driven intelligent fault diagnosis technology has received more and more attention and research. This technology establishes a fault diagnosis model based on a large amount of historical data, automatically extracts the fault information hidden in the monitoring data of running equipment, completes the intelligent identification of the health status of the equipment, and realizes more efficient pre-predictive maintenance.

[0004] Width learning combines the advantages of random vector function chain networks and feedforward neural networks, and has the characteristics of simple and fast model training and low computing resource requirements, which can meet the requirements of fast and accurate diagnosis of bearing fault types in practical applications. However, traditional width learning is generally developed based on a one-time learning method, which makes it difficult to handle the problem of new fault samples and fault modes in the data stream. Summary of the invention

[0005] The purpose of the present invention is to propose a bearing fault diagnosis method based on width incremental learning in response to the defects and improvement needs in the prior art as mentioned above, with the aim of using SVDD to perform preliminary anomaly detection on new data; at the same time, according to the anomaly detection results, width stacking is performed on width learning to achieve the sample increment or class increment effect of the model, so as to meet the high real-time and high accuracy of the model for new data.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A bearing fault diagnosis method based on width incremental learning, comprising:

[0008] Obtain the original time domain data of the bearing required for the width learning fault classification model;

[0009] Preprocessing the original time domain data of the bearing and dividing it into an old data set D1 and a new data set, wherein the new data set includes a new data set D2 and a new data set D3;

[0010] Build a width learning fault classification model and use the old data set for training to obtain the initial width learning fault classification model. Use the new data set to test the fault accuracy of the initial width learning fault classification model.

[0011] The old data set is used to train the SVDD model to obtain the initial SVDD anomaly detection model;

[0012] Use the new data set to test the SVDD anomaly detection model to determine whether the data anomaly rate reaches the anomaly detection threshold. If it is lower than the anomaly detection threshold, use width sample incremental learning; if it is higher than the anomaly detection threshold, use width class incremental learning;

[0013] The width learning fault classification model after incremental learning is used to diagnose the real-time data of the bearing and obtain the fault accuracy.

[0014] Furthermore, the original time domain data of the bearing is preprocessed and divided into old data sets and new data sets, specifically:

[0015] After DC removal, envelope analysis and downsampling processing are performed on the original time domain data of the bearing, a data set with a sampling frequency of 800 Hz is generated;

[0016] The 800Hz data set is divided into new and old data sets, where the data in the old data set D1 has multiple fault types, and all fault types in the new data set D2, i.e., the new sample incremental data set, belong to the old data set; the new data set D3, i.e., the new class incremental data set, has only one fault type that does not belong to the old data set.

[0017] Furthermore, a width learning fault classification model is built and trained with the old data set to obtain an initial width learning model. The new data set is used to test the fault accuracy of the initial width learning fault classification model. Specifically:

[0018] The initial width learning fault classification model is composed of a feature layer, an enhancement layer and an output layer;

[0019] The data in the old dataset is used as the input data of the width learning fault classification model to obtain the mapping features and enhanced features. Then, the output layer is calculated based on the mapping features and enhanced features. Combined with the label Y1 in the old dataset D1, we have:

[0020] W=(λI+AA T ) -1 A T Y1

[0021] Among them, the feature matrix A = [Z|H], W is the weight connecting the feature layer and the enhancement layer to the output layer, and Z and H represent the mapping features and enhancement features of the network respectively. After calculating W, the model training is completed;

[0022] The same operation is performed on D2 = {X2, Y2} and D3 = {X3, Y3} to obtain feature matrices A2 and A3, and further obtain predicted labels Y2′ and Y3′:

[0023] Y2′=A2W

[0024] Y3′=A3W

[0025] By comparing Y2′ and Y′3 with Y2 and Y3, the fault accuracy of the new data set D2 and the new data set D3 can be obtained.

[0026] Furthermore, the old data set is used to train the SVDD model to obtain an initial SVDD anomaly detection model, specifically:

[0027] Use the SVDD model to construct a hypersphere that can accommodate all training samples with the smallest volume. The hypersphere is represented by the center c and radius R.

[0028] The training process specifically involves solving the boundary of the old data set to obtain the boundary R;

[0029]

[0030] Where X = {x i ,i=1,2,...,I} is the old data set data. i represents the Lagrange multiplier, and K(·,·) represents the kernel function.

[0031] Furthermore, it also includes

[0032] For the samples x in the new dataset D2 and the new dataset D3 new Perform anomaly detection:

[0033]

[0034] If ||x new -a|| 2 ≤R 2 , the new data is accepted, otherwise, the new data is an outlier.

[0035] Furthermore, if the anomaly rate is lower than the anomaly detection threshold, wide sample incremental learning is used; if the anomaly rate is higher than the anomaly detection threshold, wide sample class incremental learning is used, specifically:

[0036] The width sample increment learning is to perform sample increment on the new data set D2 = {X2, Y2} data, then:

[0037]

[0038] In the formula,

[0039] UT =A2(A) + ,S T =A2-U T A

[0040]

[0041] At this point, the sample increment model is updated;

[0042] The width class incremental learning is to perform class increment on the new data set D3 = {X3, Y3};

[0043]

[0044] In the formula,

[0045] U=A + A3,S=A3-AU

[0046]

[0047] W t =A + [Y|Φ]=[W|Φ]

[0048] Among them, Φ is a zero matrix, A + It is the pseudo-inverse transformation of the input feature A.

[0049] At this point, the incremental model update is complete.

[0050] Furthermore, the width learning-based fault classification model presets the number of initial mapping feature nodes and enhanced nodes and adjusts them according to the accuracy of the fault classification model on the old data set.

[0051] Furthermore, the number of nodes k generated by the initial mapping feature is 20, the number of nodes n generated by the mapping feature is 200, and the number of nodes m generated by the enhanced feature is 200.

[0052] Furthermore, the regularization parameters of the SVDD model, the corresponding statistical features of the two-dimensional visualization, and the anomaly detection threshold are set.

[0053] Furthermore, the width learning fault classification model after incremental learning is used to diagnose the real-time data of the bearing faults, and the fault accuracy is obtained, which is:

[0054] Obtain real-time bearing vibration signal data;

[0055] The SVDD anomaly detection model is input to determine whether it is an abnormal value. If it is an abnormal value, the width learning fault classification model is incremented, and the incremented width learning fault classification model is used for fault diagnosis;

[0056] Otherwise, the width learning fault classification model is directly used for fault diagnosis.

[0057] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0058] (1) The present invention adopts the original data preprocessing operation of DC removal + envelope analysis + downsampling, which can easily remove the influence of additional signal noise and has good feature mining capabilities for the fault characteristics of bearings with frequent load and speed changes.

[0059] (2) The diagnostic model based on width incremental learning of the present invention can well handle the problem of new fault samples and fault modes appearing in the data stream. At the same time, when updating the model, width incremental learning of sample increment only needs to store the old model but not the old data; width incremental learning of class increment only needs to store the label information of the old model and the old data, which greatly reduces the pressure on computer data storage, saves costs, and has practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic diagram of the overall process of an embodiment of the method of the present invention.

[0061] Figure 2 is the data set distribution used by the method of the present invention.

[0062] Figure 3(a)-Figure 3(b) It is a network structure diagram of width learning and width incremental learning of the method of the present invention.

[0063] Figure 4(a)-Figure 4(c) They are schematic diagrams of implementing SVDD anomaly detection by the method of the present invention. DETAILED DESCRIPTION

[0064] The present invention will be further described in detail below in conjunction with examples, but the embodiments of the present invention are not limited thereto.

[0065] Example

[0066] like Figure 1 As shown, a bearing fault diagnosis method based on width incremental learning includes the following:

[0067] Obtain the original time domain data of the bearing required for the width learning fault classification model,

[0068] This embodiment uses a public laboratory data set as the source of raw data, specifically, the raw data of the Case Western Reserve University bearing data set as experimental data. The data set is required to contain bearing vibration data with complete fault types and sufficient data volume, complete and correct labels, and balanced data volume of various health statuses.

[0069] like Figure 2As shown, this embodiment uses the vibration signal of a 7mil damaged bearing under four load conditions (0, 1, 2, 3hp) at the motor drive end. The bearing health status includes: no fault, inner ring fault, outer ring fault, rolling element fault. The sampling frequency is 12000Hz.

[0070] The original time domain data of the bearing are preprocessed and divided into an old data set D1 and a new data set, wherein the new data set includes a new data set D2 and a new data set D3.

[0071] Specifically:

[0072] Perform DC removal operations on all original time domain data to extract useful AC information from bearing signals;

[0073] Perform envelope analysis on the time series obtained after DC operation to achieve the purpose of denoising and obtain the time domain signal after envelope;

[0074] Downsample the time domain signal after enveloping. The signal is uniformly reduced to 800Hz sampling frequency to facilitate subsequent rapid model training;

[0075] The 800Hz signal bearing data set is divided into the old data set D1 = {X1, Y1}, the new sample incremental data set D2 = {X2, Y2}, and the new class incremental data set D3 = {X3, Y3}. The D1 data set has sufficient data and more labels (fault classification); D2 has sufficient data and the label class belongs to D1; D3 has sufficient data and the fault data of the label class does not belong to D1.

[0076] As shown in Figure 3(a) and Figure 3(b), a width learning fault classification model is built, the number of initial mapping feature nodes and enhanced nodes is set, and the old data set is used for training to obtain the initial width learning fault classification model. The new data set is used to test the fault accuracy of the initial width learning fault classification model.

[0077] The width learning fault classification model consists of a feature layer, an enhancement layer and an output layer.

[0078] Specifically:

[0079] Construct the feature layer Z based on auto-encoder. Multiple time domain sequences in the fault dataset D1 are used as the input of the feature extraction network X = {x i ,i=1,2,...,I}, assuming that each feature node window of the width learning network has n mapping features, the feature expression of the i-th mapping is shown in formula (1):

[0080]

[0081] Where: Weight and deviation Randomly generated and optimized by a sparse autoencoder, Usually a linear transformation.

[0082] definition m=1,2,…,k is the number of nodes generated for each mapping feature. The final mapping feature is expressed as The mapping feature Z is connected to the enhancement layer through nonlinear transformation to form an enhanced feature. Assuming that the width learning network enhancement layer m has an enhancement node, the expression of the jth enhanced feature is shown in formula (2):

[0083]

[0084] Where: activation function ξ = relu(x), and Randomly generated.

[0085] The enhanced features of the enhancement layer are expressed as The mapping feature Z and the enhanced feature H are input to the output layer of the width learning network. It is known that the label of the input data X is Y. The relationship between the mapping feature Z, the enhanced feature H and the label Y is shown in formula (3):

[0086] Y=[Z|H]W (3)

[0087] Where: W is the weight connecting the feature layer and the enhancement layer to the output layer. The calculation of W is shown in formula (4):

[0088] W=[Z|H] + Y (4)

[0089] In the formula: Let the feature matrix A = [Z | H], then W = A + Y, A + represents the pseudo-inverse matrix of A.

[0090] The pseudo-inverse is the least squares estimator of the linear equation. Its goal is to solve the output weight W with the minimum training error. In particular, when the feature matrix A is not a full-rank matrix, it is impossible to solve A directly. + Therefore, the ridge regression algorithm is introduced and equation (5) is used to solve W.

[0091]

[0092] In formula (5), σ1>0, σ2>0, u, v represent norm regularization, and the optimization problem of formula (5) is set to L2 norm regularization by σ1=σ2=u=v=2.

[0093] The regularization function represented by formula (5) is a convex function with good generalization performance, where λ is a further restriction on the weight. Theoretically, if λ = 0, formula (5) evolves into a least square problem; if λ → ∞, the solution of formula (5) is strictly restricted and tends to 0. Therefore, λ is usually set to 0. Taking the derivative of formula (5) and setting the derivative equal to 0, we can get:

[0094] W=(λI+AA T ) -1 A T Y (6)

[0095] In formula (6), I represents the unit matrix. The process of obtaining W is the training process of the model (including increment and non-increment). At this point, the model training is completed.

[0096] Finally, the same operation is performed on D2 = {X2, Y2} and D3 = {X3, Y3} to obtain the feature matrices A2 and A3, and the predicted labels Y2′ and Y3′ are further calculated, so as to calculate the fault accuracy of the new data D2 and D3 in combination with the known Y2 and Y3, which is the model testing process:

[0097] Y2′=A2W (7)

[0098] Y3′=A3W. (8)

[0099] Furthermore, the number of initial mapping feature nodes and enhanced nodes is preset. The number of nodes is adjusted according to the accuracy of the model on D1. The default number of initial mapping feature nodes k is 20, the number of nodes generated by mapping features n is 200, and the number of nodes generated by enhanced features m is 200.

[0100] like Figure 4(a)-Figure 4(c) As shown, the regularization statement of the SVDD model, the corresponding statistical features of the two-dimensional visualization and the anomaly detection threshold are set, the SVDD model is trained with the old data set to obtain the initial SVDD anomaly detection model, and the anomaly rate of the new data set is further obtained.

[0101] As shown in Figure 4(b), the orange batch of data is mostly old fault type data, while the batch of data in Figure 4(c) is mostly new fault type data.

[0102] Given a training data set X = {x i ,i=1,2,...,I}. Use SVDD to construct a hypersphere that can accommodate all training samples with the minimum volume. The hypersphere can be represented by the center c and radius R, so the optimization problem of SVDD can be expressed as:

[0103]

[0104] Among them, ξ iis a slack variable used to penalize outliers that may exist in the training dataset. The regularization parameter C is used to trade off between quantity and error.

[0105] Using the Lagrange multiplier, the following quadratic optimization problem can be obtained:

[0106]

[0107] α i represents the Lagrange multiplier, K(·,·) represents the kernel function, and the Gaussian kernel is used here. The KKT conditions are shown in Equation (11):

[0108]

[0109] The few samples located on the boundary of the hypersphere are called support vectors x sv , whose α i > 0.

[0110] Calculate R 2 , and the specific calculation formula is shown in Equation (12):

[0111]

[0112] where x k ∈x sv＜C , that is, the support vector set where α k < C.

[0113] Furthermore, based on the initial SVDD anomaly detection model, the anomaly rate of the new dataset is further obtained, specifically:

[0114] To test a data x new in the new dataset, the distance to the center of the hypersphere is obtained through the following formula (13):

[0115]

[0116] If ||x new - a|| 2 ≤ R 2 , then the new data is accepted, otherwise, the new data is an outlier.

[0117] When the proportion of new data outliers in dataset D2 or D3 is less than the threshold, the model is updated using width sample incremental learning later. Otherwise, the model is updated using width class incremental learning.

[0118] In this method, judging the relationship between x new and R is the process of SVDD testing.

[0119] As shown in Figure 3(b), the SVDD anomaly detection model is tested using a new data set to determine whether the data anomaly rate reaches the anomaly detection threshold. If it is lower than the anomaly detection threshold, the threshold for this experiment is set to 85%, then width sample incremental learning is used; if it is higher than the anomaly detection threshold, then width class incremental learning is used.

[0120] Specifically:

[0121] X a new set of data new Perform SVDD outlier detection. If the outlier rate of the batch of data is greater than 85%, it is considered that X new Most of the data are of different types from the old data, and incremental learning is performed. If it is less than or equal to 85%, it is considered that X new The fault types of most of the data are among the fault types of the old data, and sample incremental learning is performed. In the incremental learning process, the weight matrix W and feature A of the old samples in the training width learning fault classification model are introduced for subsequent calculations.

[0122] In the sample increment process, the model weight parameter W is calculated. D2 Update:

[0123] X r+1 Represents X new , Z r+1 and H r+1 Respectively represent X r+1 The linear mapping features and nonlinear enhancement features of r are the fault samples, and r+1 represents the sample increment. Therefore, the new feature set A of width learning r+1 It can be written as:

[0124]

[0125] a r+1 Represents the feature set of the new fault sample, that is, a r+1 =[Z r+1 |H r+1 ]. New feature set A r+1 The pseudo-inverse value of can be obtained as follows:

[0126] (A r+1 ) + =[(A) + -VU T |V] (15)

[0127] In the formula,

[0128] U T =a r+1 (A) + ,S T =a r+1 -U T A (16)

[0129]

[0130] Finally, the new weight matrix W can be updated by r+1 .

[0131] W r+1 =W+(y r+1 -a r+1 W)V (18)

[0132] Among them, y r+1 is the fault label of the new sample.

[0133] In class increment, assume that the old data set contains m failure modes. The new data X new A new type of failure mode data appears, m represents the failure class, and m+1 represents the class increment. In order to accommodate all these failure modes, the dimension of Y should be expanded from N×m to N×(m+1). Where Φ is a zero matrix with dimension N×1. The weight W can be enlarged to W t , as shown below:

[0134]

[0135] Among them, A + It is the pseudo-inverse transformation of the input feature A. The input feature of the new fault category is recorded as h m+1 , which includes convolutional features and additional features. Then, the input features of the width incremental learning model can be expressed as Similar to (14), The calculation method is as follows:

[0136]

[0137] In the formula,

[0138] U=A + h m+1 ,S=h m+1 -AU (21)

[0139]

[0140] According to the obtained The weights of the fully connected layer can be updated as follows:

[0141]

[0142] Among them, y m+1 is the enlarged label of the newly appeared abnormal sample, whose dimension is N new×(m+1),N new Is the new data X new The number of samples in .

[0143] Using the diagnostic method of the present invention, the predicted labels of X2 and X3 in the new data set are calculated and the accuracy is obtained. The results are shown in the following table. Table 1 (a) shows the accuracy of the width learning model on D1 and D2 before and after sample increment, and Table 1 (b) shows the accuracy of the width learning model on D1 and D3 before and after class increment.

[0144] Table 1 (a) Incremental model accuracy

[0145]

[0146] Table 1(b) Accuracy of sample increment model

[0147]

[0148] The application of this method in the actual production environment is:

[0149] Obtain real-time bearing vibration signal data;

[0150] The SVDD anomaly detection model is input to determine whether it is an abnormal value. If it is an abnormal value, the width learning fault classification model is incremented, and the incremented width learning fault classification model is used for fault diagnosis;

[0151] Otherwise, the width learning fault classification model is directly used for fault diagnosis.

[0152] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A bearing fault diagnosis method based on width incremental learning, characterized in that: include: Obtain the original time domain data of the bearing required for the width learning fault classification model; Preprocessing the original time domain data of the bearing and dividing it into an old data set D1 and a new data set, wherein the new data set includes a new data set D2 and a new data set D3; Build a width learning fault classification model and use the old data set for training to obtain the initial width learning fault classification model. Use the new data set to test the fault accuracy of the initial width learning fault classification model. The old data set is used to train the SVDD model to obtain the initial SVDD anomaly detection model; Use the new data set to test the SVDD anomaly detection model to determine whether the data anomaly rate reaches the anomaly detection threshold. If it is lower than the anomaly detection threshold, use width sample incremental learning. If it is above the anomaly detection threshold, width class incremental learning is used; The width learning fault classification model after incremental learning is used to diagnose the real-time data of the bearing and obtain the fault accuracy.

2. The bearing fault diagnosis method according to claim 1, characterized in that: The original time domain data of the bearing is preprocessed and divided into old data sets and new data sets, specifically: After DC removal, envelope analysis and downsampling processing are performed on the original time domain data of the bearing, a data set with a sampling frequency of 800 Hz is generated; The 800Hz data set is divided into new and old data sets, where the data in the old data set D1 has multiple fault types, and all fault types in the new data set D2, i.e., the new sample incremental data set, belong to the old data set; the new data set D3, i.e., the new class incremental data set, has only one fault type that does not belong to the old data set.

3. The bearing fault diagnosis method according to claim 1, characterized in that: Build a width learning fault classification model and use the old data set for training to obtain the initial width learning model. Use the new data set to test the fault accuracy of the initial width learning fault classification model. Specifically: The initial width learning fault classification model is composed of a feature layer, an enhancement layer and an output layer; The data in the old dataset is used as the input data of the width learning fault classification model to obtain the mapping features and enhanced features. Then, the output layer is calculated based on the mapping features and enhanced features. Combined with the label Y1 in the old dataset D1, we have: W=(λI+AA T ) -1 A T Y1 in , feature matrix A = [Z|H], W is the weight connecting the feature layer and enhancement layer to the output layer, then the model training is completed; The same operation is performed on D2 = {X2, Y2} and D3 = {X3, Y3} to obtain feature matrices A2 and A3, and further obtain predicted labels Y′2 and Y′3: Y′2=A2W Y′3=A3W By comparing Y′2 and Y′3 with Y2 and Y3, the fault accuracy of the new data set D2 and the new data set D3 can be obtained.

4. The bearing fault diagnosis method according to claim 1, characterized in that: The old data set is used to train the SVDD model to obtain the initial SVDD anomaly detection model, which is specifically: Use the SVDD model to construct a hypersphere that can accommodate all training samples with the minimum volume. The hypersphere is represented by the center c and radius R. The training process is to solve the boundary of the old data set to obtain R.

5. The bearing fault diagnosis method according to claim 4, characterized in that: It also includes the samples x in the new data set D2 and the new data set D3 new Perform anomaly detection: If ||x new -a|| 2 ≤R 2 , the new data is accepted, otherwise, the new data is an outlier.

6. The bearing fault diagnosis method according to claim 1, characterized in that: If the anomaly rate is lower than the anomaly detection threshold, wide sample incremental learning is used; if the anomaly rate is higher than the anomaly detection threshold, wide sample class incremental learning is used, specifically: The width sample increment learning is to perform sample increment on the new data set D2 = {X2, Y2} data, then: In the formula, U T =A2(A) + ,S T =A2-U T A At this point, the sample increment model is updated; The width class incremental learning is to perform class increment on the new data set D3 = {X3, Y3}; In the formula, <h2 style=";text-align:left;direction:ltr">U=A<h2 style=";text-align:left;direction:ltr"> + <h2 style=";text-align:left;direction:ltr"> A3,S=A3-AU W t =A + [Y|Φ]=[W|Φ] At this point, the incremental model update is complete.

7. The bearing fault diagnosis method according to claim 1, characterized in that: The width learning fault classification model presets the number of initial mapping feature nodes and enhanced nodes, and is adjusted according to the accuracy of the width learning fault classification model on the old data set.

8. The bearing fault diagnosis method according to claim 7, characterized in that: The number of nodes generated by the initial mapping feature is 20, the number of nodes generated by the mapped feature is 200, and the number of nodes generated by the enhanced feature is 200.

9. The bearing fault diagnosis method according to claim 1, characterized in that: Both wide sample incremental learning and wide sample class incremental learning perform wide stacking on the feature layer of the wide learning fault classification model.

10. The bearing fault diagnosis method according to any one of claims 1 to 9, characterized in that: The width learning fault classification model after incremental learning is used to diagnose the real-time data of the bearing and obtain the fault accuracy, which is as follows: Obtain real-time bearing vibration signal data; The SVDD anomaly detection model is input to determine whether it is an abnormal value. If it is an abnormal value, the width learning fault classification model is incremented, and the incremented width learning fault classification model is used for fault diagnosis; Otherwise, the width learning fault classification model is directly used for fault diagnosis.

Images