Fault diagnosis model, method, system and device based on small sample learning and readable medium

Abstract

The invention relates to the technical field of fault diagnosis, in particular to a fault diagnosis model, method, system and device based on small sample learning and a readable medium. The invention provides a construction method of a fault diagnosis model based on small sample learning, which comprises the following steps: firstly, collecting vibration signal data on a target moving part, and converting the vibration signal data into a feature image to construct a data set in which the feature image marked with a fault category is stored; performing machine learning on the constructed fault diagnosis model on a data set; the fault diagnosis model comprises a feature extraction module and a classification module, the feature extraction module learns a deep local descriptor of a feature image based on a residual element multiplication network to output a local feature, and the classification module queries the similarity between the feature image and each fault category by adopting a proximity classification algorithm based on the output of the feature extraction module. According to the method provided by the invention, the fault diagnosis effect can be remarkably improved, and the accuracy of the diagnosis result when the training sample is unbalanced under the small sample is improved.

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis technology, and in particular to a fault diagnosis model, method, system, device and readable medium based on small sample learning. Background Art

[0002] Traditional bearing fault diagnosis methods usually rely on large-scale historical data and accurate models to predict and identify possible failure modes. However, in some cases, data may be limited. For example, industrial data often has a high degree of category deviation. Most of the time, the equipment is in normal operation, the number of fault samples is far less than the number of healthy samples, and only limited sample data can be used for analysis. In this case, traditional machine learning methods often have difficulty accurately identifying potential failure modes and predicting possible problems with bearings in advance.

[0003] The technology based on small sample learning has proposed a new solution to this problem. The key advantage of small sample learning is that it can build an efficient model when data is scarce. This method uses a small number of data samples for model training, can use limited data and extract key features from it, and uses special algorithms and techniques to effectively identify potential failure modes of bearings. Then identify potential failure signals of bearings. In addition, this method also has good generalization ability, can adapt to new data and continuously improve the accuracy of the model to learn and infer a wider range of patterns and features from a small number of samples, thereby improving the reliability and accuracy of bearing fault diagnosis. . Summary of the invention

[0004] In order to overcome the defect in the above-mentioned prior art that bearing diagnosis is difficult to adopt machine learning due to lack of samples, the present invention proposes a method for constructing a fault diagnosis model based on small sample learning, which can construct a high-precision fault diagnosis model based on a small number of healthy samples. This method is suitable for mechanical moving parts such as bearings.

[0005] The present invention proposes a method for constructing a fault diagnosis model based on small sample learning. First, vibration signal data is collected on a target moving part, and the vibration signal data is converted into a feature image to construct a data set storing the feature image with marked fault categories; then the constructed fault diagnosis model is made to perform machine learning on the data set;

[0006] The fault diagnosis model includes a feature extraction module and a classification module. The feature extraction module learns the deep local descriptor of the feature image based on the residual element multiplication network to output local features. The classification module uses a neighbor classification algorithm based on the output of the feature extraction module to query the similarity between the feature image and each fault category.

[0007] Preferably, the feature extraction module includes a plurality of residual blocks connected in sequence; the residual block includes: a first activation function, a batch normalization layer, a convolution layer, a first fully connected layer, a second fully connected layer, a third fully connected layer, a dimensionality superposition unit and a second activation function; the first activation function, the batch normalization layer and the convolution layer are connected in sequence, the input of the first fully connected layer and the input of the second fully connected layer are both connected to the output of the convolution layer, and the output of the first fully connected layer and the output of the second fully connected layer are both connected to the input of the third fully connected layer; the input of the dimensionality superposition unit is respectively connected to the input of the fault diagnosis model and the output of the third fully connected layer, the output of the dimensionality superposition unit is connected to the input of the second activation function, the dimensionality superposition unit performs dimensionality superposition on the feature image of the fault diagnosis model input and the output data of the third fully connected layer and then sends them to the second activation function for activation, and the output of the second activation function is the output of the residual block.

[0008] Preferably, the classification module embeds the local features output by the feature extraction module into the vector space, and calculates the similarity of each feature image in the vector space. The classification module combines the similarity to find k nearest neighbors of the feature image of the sample to be detected from each fault category, and calculates the probability that the sample to be detected belongs to each fault category based on the neighbor similarity, where k is a set value.

[0009] Preferably, the calculation formula for the probability p(i) that the sample to be detected belongs to the i-th category is as follows:

[0010]

[0011] Where M is the dimension of the embedding vector space, x m The mth array representing the local features ψ(X) of the sample to be detected after being processed by the feature extraction module, 1≤m≤M; Represents the local features of the jth sample in the ith category after being processed by the feature extraction module The m-th array of , the superscript T indicates transpose; ||.|| indicates the bi-norm.

[0012] Preferably, during the iteration process of the fault diagnosis model, the loss function Loss is calculated according to the following formula:

[0013]

[0014] Among them, y(i) is the parameter of the true category label probability distribution after smoothing, ε is the set smoothing parameter, and ε takes a value between 0 and 1; n is the number of fault categories; p(i) is the probability that the verification sample predicted by the fault category diagnosis model belongs to the i-th fault category.

[0015] Preferably, the two-dimensional image after the vibration signal data is converted is recorded as the original image, and the feature image in the data set includes the original image and the image obtained by performing image enhancement on the original image.

[0016] A fault diagnosis method proposed in the present invention first adopts the method for constructing a fault diagnosis model based on small sample learning to construct a fault diagnosis model for the object to be diagnosed; then, vibration signal data of the object to be diagnosed is collected and converted into a two-dimensional image, and the two-dimensional image is input into the fault diagnosis model as a sample to be detected, and the fault category corresponding to the maximum probability output by the fault diagnosis model is obtained as a diagnosis result.

[0017] A fault diagnosis device proposed in the present invention includes: a signal acquisition module, a signal processing module and a fault diagnosis model; the signal acquisition module is used to collect vibration signal data of an object to be diagnosed; the signal processing module is respectively connected to the signal acquisition module and the fault diagnosis model, the signal acquisition module converts the vibration signal data into a two-dimensional image and then inputs it into the fault diagnosis model, and the fault diagnosis model outputs a diagnosis result.

[0018] A fault diagnosis system proposed in the present invention includes a memory and a processor, wherein a computer program is stored in the memory, and the processor is connected to the memory, and the processor is used to execute the computer program to implement the fault diagnosis method.

[0019] The present invention provides a readable medium storing a computer program, which is used to implement the fault diagnosis method when executed.

[0020] The advantages of the present invention are:

[0021] The present invention converts one-dimensional raw vibration data into two-dimensional images as the input of the model, and uses metric learning deep learning theory to construct a feature extraction module and a classification module based on residual element multiplication, which amplifies the feature dimension and reduces the computational overhead in the process of feature extraction. At the same time, y(i) is introduced, and the label smoothing cross entropy loss function is used to update the model parameters to reduce excessive dependence on training data, thereby reducing the adverse effects of imbalanced training samples and improving generalization ability, and improving the accuracy of fault diagnosis under small samples. The method proposed in the present invention can significantly improve the fault diagnosis effect and improve the accuracy of the diagnosis results when the training samples are unbalanced under small samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is the training flow chart of the bearing operation fault diagnosis model based on small sample learning

[0023] Figure 2 A simulation diagram showing the comparison of the training accuracy obtained by the method of the present invention and the comparative method in the embodiment

[0024] Figure 3 Schematic diagram showing the comparison of loss values ​​obtained by the method of the present invention and the comparative method in the embodiment

[0025] Figure 4 A simulation diagram showing the comparison of the test accuracy obtained by the method of the present invention and the comparative method in the embodiment

[0026] Figure 5 Schematic diagram showing the comparison of verification accuracy obtained by the method of the present invention and the comparative method in the embodiment

[0027] ResEMNet-LabelSmooth, ResEMNet--CrossEntropy, Conv64F-CrossEntropy, Conv64F-LabelSmooth. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] The data samples in this embodiment are recorded as {feature image, fault category}; the feature image includes two types, one is a two-dimensional image obtained by converting the sampled vibration data signal through signal processing, and the other is an image obtained by enhancing the first type of two-dimensional image;

[0030] The vibration data signal can be processed by the matplotlib library and converted into a two-dimensional image.

[0031] Image enhancement includes: random photometric transformation, random cropping, etc.

[0032] In this embodiment, the fault diagnosis model performs fault diagnosis on the characteristic image obtained based on the vibration data signal processing and outputs the fault category.

[0033] The fault diagnosis model includes: a feature extraction module and a classification module; the feature extraction module includes a first residual block, a second residual block, a third residual block and a fourth residual block connected in sequence;

[0034] The first residual block, the second residual block, the third residual block and the fourth residual block have the same structure and are collectively referred to as residual blocks. The input of the residual module is used as the input of the fault diagnosis model.

[0035] The residual block includes: a first activation function, a batch normalization layer, a convolution layer, a first fully connected layer, a second fully connected layer, a third fully connected layer, a dimension stacking unit, and a second activation function. The first activation function, the batch normalization layer, and the convolution layer are sequentially connected, and the feature image is activated by the first activation function, and then output after batch normalization and convolution processing.

[0036] The input of the first fully connected layer and the input of the second fully connected layer are both connected to the output of the convolution layer, and the output of the first fully connected layer and the output of the second fully connected layer are both connected to the input of the third fully connected layer. In this way, the output of the convolution layer is processed by the first fully connected layer and the second fully connected layer respectively, and then the output of the second fully connected layer and the output of the second fully connected layer are processed by point multiplication and then input to the third fully connected layer for processing. The input of the dimension superposition unit is respectively connected to the input of the fault diagnosis model and the output of the third fully connected layer, and the output of the dimension superposition unit is connected to the input of the second activation function. The dimension superposition unit performs dimension superposition on the feature image input of the fault diagnosis model and the output data of the third fully connected layer and then sends them to the second activation function for activation. The output of the second activation function is the output of the residual block.

[0037] In this way, let the input of the residual block be recorded as input and the output be recorded as output, then the processing formula of the residual block is expressed as follows:

[0038] DW=Conv(BN(LR1(input)))

[0039] output=LR2(input+FC3(FC1(DW)*FC2(DW)))

[0040] Among them, DW represents the output of the convolutional layer, Conv represents the convolutional layer, BN represents the batch normalization layer, LR1 represents the first activation function, LR2 represents the second activation function, FC1, FC2 and FC3 represent the first fully connected layer, the second fully connected layer and the third fully connected layer respectively.

[0041] The first residual block, the second residual block, the third residual block and the fourth residual block are connected in sequence, and the output of the fourth residual block is the output of the feature extraction module. Let the first residual block, the second residual block, the third residual block and the fourth residual block be denoted as B1, B2, B3 and B4 respectively, and the feature image input to the fault diagnosis model be denoted as X', then the local feature ψ(X') output by the feature extraction module is expressed as follows:

[0042] ψ(X')=B4(B3(B2(B1(X'))))

[0043] Classification module

[0044] The classification module first embeds the local features output by the feature extraction module into the vector space Rd×M , written as:

[0045] ψ(X')=[x'1,x'2,…,x' m ,…,x' M ]

[0046] x' m The mth array representing the local feature ψ(X'), 1≤m≤M; each array contains d values; that is, x'1, x'2, x' M They represent the first, second and Mth arrays of the local features ψ(X') respectively.

[0047] The classification module combines the similarity to find the k nearest neighbors of the sample to be detected from each fault category, and calculates the probability that the sample to be detected belongs to each fault category based on the neighbor similarity.

[0048] Taking the i-th fault category as an example, the classification module first calculates the similarity between the sample to be detected and each sample in the i-th fault category in the data set, and then selects the k samples with the largest similarity as the neighbors of the sample to be detected. k is a set value. In specific implementation, any integer greater than 1 can be selected, such as 3, 5, 10, etc.

[0049] Let the sample to be detected be recorded as X. After being processed by the feature extraction module, we have:

[0050] ψ(X)=[x1,x2,…,x m ,…,x M ]

[0051] x m The mth array representing the local feature ψ(X), 1≤m≤M; each array contains d values; that is, x1, x2, x M They represent the first, second and Mth arrays of the local features ψ(X) respectively.

[0052] Let the jth sample of the i-th fault category in the data set be denoted as After being processed by the feature extraction module, we have:

[0053]

[0054] Representing local features The mth array of , 1≤m≤M; each number contains d values; that is, Represent local features The 1st array, the 2nd array and the Mth array.

[0055] Let the sample to be tested X and the sample The similarity is expressed by cosine similarity, then:

[0056]

[0057] The samples in the i-th fault category are sorted in descending order according to the cosine similarity, and the probability p(i) that the sample to be detected belongs to the i-th fault category is:

[0058]

[0059] Among them, Φ represents the classification module, c i represents the i-th fault category;

[0060] The output of the classification module is the probability {p(i)} that the sample to be tested belongs to each fault category 1≤i≤n , n is the total number of fault categories

[0061] In this implementation, the training of the fault diagnosis model includes the following steps:

[0062] S1. Extract data samples {feature image, fault category} from the data set to construct training set and validation set respectively; build a fault diagnosis model;

[0063] S2, extract training samples from the training set, and let the fault diagnosis model learn the training samples;

[0064] S3, extract verification samples from the verification set, let the fault diagnosis model predict the fault category probability of the verification samples, and calculate the loss function Loss;

[0065]

[0066] Among them, y(i) is the parameter of the true category label probability distribution after smoothing, ε is the set smoothing parameter, and ε takes a value between 0 and 1; n is the number of fault categories; p(i) is the probability that the verification sample predicted by the fault category diagnosis model belongs to the i-th fault category;

[0067] S4, judging whether the fault diagnosis model has converged; if not, returning to step S2; if yes, fixing the fault diagnosis model and outputting it.

[0068] A bearing fault diagnosis method proposed in this embodiment includes the following steps:

[0069] St1. Collect the vibration signal data of the bearing in a continuous time, perform a two-dimensional transformation on the vibration signal data, and obtain a two-dimensional image as a sample to be detected.

[0070] It is worth noting that the length of the vibration signal data is consistent during the model training process and the application process.

[0071] St2. Input the sample to be detected into the trained fault diagnosis model, and obtain the fault category corresponding to the maximum probability output by the fault diagnosis model as the diagnosis result.

[0072] In combination with a specific embodiment, the above fault prediction model is verified in bearing diagnosis. In this embodiment, the vibration signal is sampled when the bearing is in working condition, and a vibration signal data contains 6003 columns, wherein the first column is the id field representing the sample number, the 2nd to 6001st columns are the vibration signal values ​​sampled continuously in time series, and the last column label field is the label data, i.e. the fault category of the bearing. In this embodiment, the fault category is represented by numbers 0 to 9, where "0" represents normal, and other labels are shown in Table 1.

[0073] Table 1 Bearing fault status label

[0074]

[0075] In Table 1, the number "1" indicates that the bearing uses diameter A1 and is in outer ring fault state during sampling; and so on.

[0076] In this embodiment, a total of 792 vibration signal data are collected when the bearing is working, including all labels from 0 to 9. The vibration signal data is converted into a two-dimensional image after being processed by the matplotlib library, and then a data set is constructed. The data samples in the data set are recorded as {feature image, fault category}. The feature image is a two-dimensional image after the vibration signal data is converted, or the image of the two-dimensional image after image enhancement; the fault category in the data sample is the working state of the vibration signal data corresponding to the feature image when it is sampled, which is represented by numbers 0-9 in this embodiment. In this embodiment, the data set is divided into a training set, a validation set, and a test set in proportions of 60%, 20%, and 20%.

[0077] In this embodiment, the fault diagnosis model proposed in the present invention is recorded as ResEMNet-LabelSmooth; in addition, three comparison models are selected, namely:

[0078] ResEMNet--CrossEntropy: adopts the fault diagnosis model architecture proposed in the present invention. During the model training process, the loss function adopts the cross entropy loss function;

[0079] Conv64F-CrossEntropy: Compared with the fault diagnosis model architecture proposed in the present invention, the feature extraction module is replaced by a standard convolutional layer architecture, and during the model training process, the loss function adopts the cross entropy loss function;

[0080] Conv64F-LabelSmooth: Compared with the fault diagnosis model architecture proposed in the present invention, the feature extraction module is replaced with a standard convolutional layer architecture, and during the model training process, the loss function adopts the traditional label smoothing loss.

[0081] All four models were trained on the training set and validation set, and the model performance was tested on the test set after the model converged.

[0082] In this embodiment, when the four models are learned on the training set, the training parameters are set to 5-way one-shot, that is, there are 5 categories, each category has only one sample, the batchsize (batch size per training round) is set to 16, the total epoch (total number of training rounds) is 50, and the episode (number of model iterations, i.e., the number of loss transmissions) is 100; that is, in each iteration, 16 batches of 80 samples are first used for training, each batch uses 5 samples from 5 different categories; then the validation set samples are extracted to calculate the loss, and the model parameters are updated according to the loss; in this embodiment, each 100 iterations of the model are counted as one round of training, and the model convergence condition is: completing 50 rounds of training.

[0083] In this embodiment, the prediction accuracy of the four models on the training set is as follows: Figure 2 As shown, it can be seen that the feature extraction module combined with the label smoothing regularization method adopted by the model ResEMNet-LabelSmooth proposed in the present invention has the highest prediction accuracy on the training set compared to the traditional convolutional network architecture of the model Conv64F-CrossEntropy and the original cross entropy loss. At the same time, compared with the model Conv64F-LabelSmooth, the model ResEMNet-LabelSmooth proposed in the present invention proves that the feature extraction module can converge faster than the traditional convolutional network architecture during the training process.

[0084] Figure 3 This is a comparison of the loss values ​​of the four models during the training process. It can be seen that even after the label smoothing regularization method is adopted, the loss in the present invention can still converge quickly.

[0085] Figure 4This is a comparison of the prediction accuracy of the four models on the test set. It can be seen that after the four models converge, the prediction accuracy can reach more than 90%, but the model ResEMNet-LabelSmooth proposed by the present invention and the comparison model ResEMNet--CrossEntropy using the feature extraction module can converge faster, and the prediction accuracy of the model ResEMNet-LabelSmooth proposed by the present invention is smoother and more stable than that of ResEMNet--CrossEntropy. It can be seen that the model constructed by the present invention is more stable, and the test accuracy will not decrease with the increase of iteration rounds, avoiding overfitting. It can be seen that the present invention has better convergence stability and good robustness.

[0086] Figure 5 This is a comparison of the prediction accuracy of the four models on the validation set. It can be seen that the method proposed in the present invention performs best on the validation set compared with the comparison model. Figure 2 , Figure 3 , Figure 4 , Figure 5 It can be seen that the model proposed in the present invention has the highest accuracy in the training set, validation set and test set compared with other comparison models, and the model converges faster and has more stable performance.

[0087] Of course, it is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any reference numerals in the claims should not be regarded as limiting the claims involved.

[0088] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

[0089] The techniques, shapes, and structural parts not described in detail in the present invention are all well-known techniques.

Claims

1. A method for constructing a fault diagnosis model based on small sample learning, characterized in that: First, vibration signal data is collected on the target moving parts, and the vibration signal data is converted into feature images to construct a data set storing feature images with marked fault categories; then the constructed fault diagnosis model is used for machine learning on the data set; The fault diagnosis model includes a feature extraction module and a classification module. The feature extraction module learns the deep local descriptor of the feature image based on the residual element multiplication network to output local features. The classification module uses a neighbor classification algorithm based on the output of the feature extraction module to query the similarity between the feature image and each fault category.

2. The method for constructing a fault diagnosis model based on small sample learning according to claim 1, characterized in that: The feature extraction module includes a plurality of residual blocks connected in sequence; the residual block includes: a first activation function, a batch normalization layer, a convolution layer, a first fully connected layer, a second fully connected layer, a third fully connected layer, a dimensionality superposition unit and a second activation function; the first activation function, the batch normalization layer and the convolution layer are connected in sequence, the input of the first fully connected layer and the input of the second fully connected layer are both connected to the output of the convolution layer, the output of the first fully connected layer and the output of the second fully connected layer are both connected to the input of the third fully connected layer; the input of the dimensionality superposition unit is respectively connected to the input of the fault diagnosis model and the output of the third fully connected layer, the output of the dimensionality superposition unit is connected to the input of the second activation function, the dimensionality superposition unit performs dimensionality superposition on the feature image input of the fault diagnosis model and the output data of the third fully connected layer and then sends them to the second activation function for activation, and the output of the second activation function is the output of the residual block.

3. The method for constructing a fault diagnosis model based on small sample learning according to claim 1, characterized in that: The classification module embeds the local features output by the feature extraction module into the vector space and calculates the similarity of each feature image in the vector space. The classification module combines the similarity to find the k nearest neighbors of the feature image of the sample to be detected from each fault category, and calculates the probability that the sample to be detected belongs to each fault category based on the neighbor similarity, where k is a set value.

4. The method for constructing a fault diagnosis model based on small sample learning as claimed in claim 3, characterized in that: The calculation formula for the probability p(i) that the sample to be tested belongs to the i-th category is as follows: Where M is the dimension of the embedding vector space, x m The mth array representing the local features ψ(X) of the sample to be detected after being processed by the feature extraction module, 1≤m≤M; Represents the local features of the jth sample in the ith category after being processed by the feature extraction module The m-th array of , the superscript T indicates transpose; ||.|| indicates the bi-norm.

5. The method for constructing a fault diagnosis model based on small sample learning as claimed in claim 3, characterized in that: During the iteration of the fault diagnosis model, the loss function Loss is calculated according to the following formula: Among them, y(i) is the parameter of the true category label probability distribution after smoothing, ε is the set smoothing parameter, and ε takes a value between 0 and 1; n is the number of fault categories; p(i) is the probability that the verification sample predicted by the fault category diagnosis model belongs to the i-th fault category.

6. The method for constructing a fault diagnosis model based on small sample learning according to claim 1, characterized in that: The two-dimensional image after the vibration signal data is converted is recorded as the original image. The feature images in the data set include the original image and the image obtained by image enhancement of the original image.

7. A fault diagnosis method using the method for constructing a fault diagnosis model based on small sample learning as described in any one of claims 1 to 6, characterized in that: First, a fault diagnosis model of the object to be diagnosed is constructed by using the method for constructing a fault diagnosis model based on small sample learning as described in any one of claims 1 to 6; then, vibration signal data of the object to be diagnosed is collected and converted into a two-dimensional image, and the two-dimensional image is input into the fault diagnosis model as a sample to be detected, and the fault category corresponding to the maximum probability output by the fault diagnosis model is obtained as a diagnosis result.

8. A fault diagnosis device using the fault diagnosis method according to claim 7, characterized in that: include: Signal acquisition module, signal processing module and fault diagnosis model; the signal acquisition module is used to collect vibration signal data of the object to be diagnosed; the signal processing module is connected to the signal acquisition module and the fault diagnosis model respectively, the signal acquisition module converts the vibration signal data into a two-dimensional image and then inputs it into the fault diagnosis model, and the fault diagnosis model outputs the diagnosis result.

9. A fault diagnosis system, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, the processor is connected to the memory, and the processor is used to execute the computer program to implement the fault diagnosis method as claimed in claim 7.

10. A readable medium, characterized in that: A computer program is stored, and when the computer program is executed, it is used to implement the fault diagnosis method as claimed in claim 7.

Images