Bearing fault diagnosis and evaluation method based on knowledge feature fusion

Abstract

The invention relates to the technical field of bearing fault diagnosis, in particular to a bearing fault diagnosis evaluation method based on knowledge feature fusion. The method comprises the following steps: extracting historical fault feature data; training a fault diagnosis model; introducing a central loss function to train an embedded layer network model according to the output of the training data in a fault diagnosis model feature layer; mapping the model feature data by using an embedded layer network model; performing weighted fusion on the mapping data and the knowledge features based on the data; calculating a local density distribution condition of adjacent data; and calculating the confidence of the diagnosis result. According to the method, the characteristics extracted in the diagnosis process of the fault diagnosis model and the characteristics of the fault data are fused, the confidence degree condition of the model diagnosis result is estimated in the embedding space according to the sample local density distribution, the problems that the confidence degree of the fault diagnosis model is difficult to assess, and the interpretability of the general assessment process is poor are solved, and the reliability of the fault diagnosis model is improved. And a confidence evaluation result under a complex model and complex data can be obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing fault diagnosis, and in particular to a bearing fault diagnosis and evaluation method based on knowledge feature fusion. Background Art

[0002] With the rapid advancement of science and technology and the rise of modern industry, industrial equipment needs to maintain reliable performance in a variety of production environments, and is sometimes even required to operate under extreme conditions. The working conditions of many parts that make up the mechanical device are often complex and harsh. Accurate fault diagnosis of each component is one of the key tasks to ensure the smooth progress of industrial production. Traditional fault diagnosis technology mainly relies on time domain, frequency domain and time-frequency domain analysis of signal processing. Time domain analysis can evaluate the health of working parts by analyzing the time series of signals. Frequency domain analysis focuses on vibration signals and identifies the characteristic frequency of bearing faults by analyzing the frequency distribution of signals, thereby realizing fault detection and diagnosis. Although these methods have achieved certain results in fault diagnosis tasks, they have certain limitations in processing complex data. Traditional methods rely too much on expert experience and prior knowledge, which not only increases the complexity of analysis, but also limits the universality and accuracy of diagnostic methods.

[0003] In order to overcome these limitations, deep learning technology has been widely used in the field of fault diagnosis in recent years. Deep learning can automatically learn and extract features from large amounts of data without manual intervention and complex feature selection processes. This method uses deep neural networks, such as convolutional neural networks and recurrent neural networks, to identify bearing failure modes by training learned deep features, greatly improving the accuracy and efficiency of fault detection. However, one of the difficulties is the confidence assessment of deep learning diagnostic results. Due to the large number of parameters, complex structure, and poor interpretability of deep learning models, it is very difficult to conduct confidence assessment of diagnostic results from a mechanism level. Summary of the invention

[0004] The present invention discloses a bearing fault diagnosis and evaluation method based on knowledge feature fusion, and the specific method is as follows:

[0005] Obtain historical bearing fault data and extract historical fault feature data;

[0006] The fault diagnosis model is trained with historical fault feature data to obtain a fault diagnosis model;

[0007] Obtain the output of the previous layer of the output layer of the fault diagnosis model as the feature data of the fault diagnosis model;

[0008] On the basis of the fault diagnosis model, an embedding layer network model is added, and the central loss is introduced to train the embedding layer network model, so as to map the characteristic data of the fault diagnosis model with the embedding layer network model;

[0009] Weighted fusion of mapping data and data-based knowledge features;

[0010] Search for the K nearest neighbor samples of each fusion feature;

[0011] According to the search results of the neighboring samples, the local density scores of different categories of its neighboring samples are calculated;

[0012] Weighted local density is used to calculate the overall confidence of the fault diagnosis model.

[0013] Furthermore, the historical fault characteristic data includes: mean, absolute mean, effective value, average power, root mean square amplitude, peak value, peak-to-peak value, variance, skewness, kurtosis, waveform, peak value, pulse, margin, skewness and kurtosis indicators.

[0014] Furthermore, when the center loss is introduced to train the embedding layer network model, the calculation formula of the center loss is as follows:

[0015]

[0016] In the formula, M represents the number of samples in each training mini-batch, L SO represents the cross entropy loss, L Cen represents the center loss; h ci For sample c i Center, c i For x i The label of the sample category, λ is a hyperparameter, x i ∈R D , D is the dimension of the feature vector;

[0017] The calculation formula of cross entropy loss is:

[0018]

[0019] In the formula, y i is the true label, is the model predicted output probability, and log is the natural logarithm.

[0020] Furthermore, the K nearest neighbor samples of each fusion feature are searched. The specific method is as follows:

[0021] Select Euclidean distance as the distance metric between samples. i ,x j ), the corresponding distance metric calculation formula is as follows:

[0022]

[0023] In the formula, x ik and x jk is the value of the sample on the kth feature;

[0024] For each test sample fusion feature, calculate its distance metric with all training samples;

[0025] According to the calculated distance, the training samples are sorted from small to large, and then the K samples with the smallest distance are selected. These samples are the K nearest neighbor samples searched.

[0026] Furthermore, the local density scores of different categories of its neighboring samples are calculated. The specific formula is:

[0027]

[0028]

[0029] Where: y represents the test sample y i The predicted value, c represents the classification result of the model's final fault diagnosis, j = 1, 2, 3, ..., k represents the k nearest neighbor training set samples, and f (·) represents the trained model. According to the above formula, A local density score ranging from 0 to 1 will be obtained.

[0030] Furthermore, the overall confidence of the fault diagnosis model is calculated. The specific formula is as follows:

[0031]

[0032] Where: max(·) represents the score of the test set samples with the highest confidence obtained after the fusion features of the test samples are mapped through the embedding network, C model Represents the confidence score of the fault diagnosis model.

[0033] Due to the adoption of the above technical solution, the present invention has the following beneficial effects:

[0034] When performing confidence assessment on a fault diagnosis model with a certain degree of complexity, the present invention can start from the data perspective, fully consider the mapping transformation process of the model to the data and the characteristics of the data set itself, and associate the test set data and the training data of the model by calculating the local density score, so as to accurately and quantitatively assess the diagnostic results of the model and the overall confidence of the model.

[0035] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings of the present invention are as follows.

[0037] Figure 1 The figure is an overall flow chart of the method of the present invention.

[0038] Figure 2 It is a schematic diagram of the structure of the fault diagnosis model used in the implementation of the present invention.

[0039] Figure 3 This is a curve diagram of the loss function change in the fault diagnosis model training process in the implementation of the present invention.

[0040] Figure 4 This is a curve diagram of the accuracy change during the fault diagnosis model training process in the implementation of the present invention.

[0041] Figure 5 This is a graph showing the change in the loss function of the embedded network training process in the implementation of the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0043] This embodiment uses the open source Western Reserve University bearing fault diagnosis data set. The basic information of the data set is shown in Table 1, and the calculated fault data features are shown in Table 2.

[0044] Table 1 Basic information of the dataset

[0045]

[0046] Table 2 Fault data characteristic calculation table

[0047]

[0048] This embodiment is aimed at the most commonly used fault diagnosis model based on convolutional neural network in the field of fault diagnosis. The model structure diagram is as follows: Figure 2 As shown in Figure 3, it includes convolutional layer, pooling layer, flattening layer, fully connected layer, Dropout layer, and output layer. The specific model information is shown in Table 3. The model is trained using the training set, and the training effect of the model is evaluated using the test set. The loss function value transformation curve during the training process is shown in Figure 3. Figure 3 As shown, the accuracy change curve is as follows Figure 4 As shown;

[0049] Table 3 Fault diagnosis model parameters

[0050]

[0051] A bearing fault diagnosis and evaluation method based on knowledge feature fusion, such as Figure 1 As shown, the specific steps are as follows:

[0052] S1. Obtain historical bearing fault data and extract historical fault feature data, as shown in Table 2.

[0053] S2. Train the fault diagnosis model with historical fault feature data to obtain the fault diagnosis model. The model structure is as follows: Figure 2 The specific parameters are shown in Table 3.

[0054] S3. Output the fully connected layer before the output layer of the fault diagnosis model to obtain the training set and test set features extracted by the model.

[0055] S4. Input the features of the training set and the test set extracted by the model into the embedding layer for training, and obtain the mapping expression of the training set and the test set in the embedding space. The aforementioned center loss is introduced during the training process to obtain a better data mapping representation. The loss function value change curve during the training process is shown in the figure below: Figure 5 shown.

[0056] S5. After obtaining the feature mapping expression of the training set and the test set extracted by the model, it is fused with the data features obtained based on knowledge calculation. In order to fully reflect the decision-making process of the model in fault diagnosis, the feature weight extracted by the model is set to 0.8 and the data feature weight is set to 0.2 when setting the fusion weight, thereby obtaining the fusion feature.

[0057] S6. Use the KNN algorithm to search for the K nearest neighbor training samples of each test set sample fusion feature, set the parameter K=15, and record the category of the nearest neighbor samples; taking the test set sample numbered 1 as an example, the nearest neighbor sample information after the search is shown in Table 4.

[0058] Table 4. Nearest neighbor search results for test samples

[0059]

[0060] S7. For each mapped test set feature data, calculate the local density scores of different categories of its neighboring training samples according to the neighboring sample search results. The calculation results are shown in Table 5.

[0061] Table 5 Calculation results of local density fraction

[0062]

[0063] S8. Perform weighted calculation on the local density scores of all test samples estimated by the test set to obtain the overall confidence C of the fault diagnosis model. model , the result is as follows:

[0064] C model =0.95632

[0065] The experimental results show that when the present invention performs confidence assessment on a fault diagnosis model with a certain complexity, it can fully consider the mapping transformation process of the model to the data and the characteristics of the data set itself from the data perspective, and associate the test set data and the training data of the model by calculating the local density score, so as to accurately and quantitatively evaluate the diagnostic results of the model and the overall confidence of the model. After the model in the embodiment is evaluated, the overall confidence level of the model reaches 95.632%, which is consistent with the high accuracy of the model on the verification set, proving the rationality of the present invention.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A bearing fault diagnosis and evaluation method based on knowledge feature fusion, characterized in that: The specific method is as follows: Obtain historical bearing fault data and extract historical fault feature data; The fault diagnosis model is trained with historical fault feature data to obtain a fault diagnosis model; Obtain the output of the previous layer of the output layer of the fault diagnosis model as the feature data of the fault diagnosis model; On the basis of the fault diagnosis model, an embedding layer network model is added, and the central loss is introduced to train the embedding layer network model, so as to map the characteristic data of the fault diagnosis model with the embedding layer network model; Weighted fusion of mapping data and data-based knowledge features; Search for the K nearest neighbor samples of each fusion feature; According to the search results of the neighboring samples, the local density scores of different categories of its neighboring samples are calculated; Weighted local density is used to calculate the overall confidence of the fault diagnosis model.

2. The bearing fault diagnosis and evaluation method based on knowledge feature fusion according to claim 1 is characterized in that: The historical fault characteristic data includes: mean, absolute mean, effective value, average power, root mean square amplitude, peak value, peak-to-peak value, variance, skewness, kurtosis, waveform, peak value, pulse, margin, skewness and kurtosis indicators.

3. The bearing fault diagnosis and evaluation method based on knowledge feature fusion according to claim 1 is characterized in that: When the center loss is introduced to train the embedding layer network model, the calculation formula of the center loss is as follows: In the formula, M represents the number of samples in each training mini-batch, L SO represents the cross entropy loss, L Cen represents the center loss; h ci For sample c i Center, c i For x i The label of the sample category, λ is a hyperparameter, x i ∈R D , D is the dimension of the feature vector; The calculation formula of cross entropy loss is: In the formula, y i is the true label, is the model predicted output probability, and log is the natural logarithm.

4. The bearing fault diagnosis and evaluation method based on knowledge feature fusion according to claim 3 is characterized in that: Search for the K nearest neighbor samples of each fusion feature. The specific method is as follows: Select Euclidean distance as the distance metric between samples. i ,x j ), the corresponding distance metric calculation formula is as follows: In the formula, x ik and x jk is the value of the sample on the kth feature; For each test sample fusion feature, calculate its distance metric with all training samples; According to the calculated distance, the training samples are sorted from small to large, and then the K samples with the smallest distance are selected. These samples are the K nearest neighbor samples searched.

5. The bearing fault diagnosis and evaluation method based on knowledge feature fusion according to claim 4 is characterized in that: Calculate the local density scores of different categories of its neighboring samples. The specific formula is: Where: y represents the test sample y i The predicted value, c represents the classification result of the model's final fault diagnosis, j = 1, 2, 3, ..., k represents the k nearest neighbor training set samples, and f (·) represents the trained model. According to the above formula, A local density score ranging from 0 to 1 will be obtained.

6. The bearing fault diagnosis and evaluation method based on knowledge feature fusion according to claim 5 is characterized in that: Calculate the overall confidence of the fault diagnosis model. The specific formula is as follows: Where: max(·) represents the score of the test set samples with the highest confidence obtained after the fusion features of the test samples are mapped through the embedding network, C model Represents the confidence score of the fault diagnosis model.

Images