A fault diagnosis method based on Euler feature extraction

Abstract

The application discloses a fault diagnosis method based on Euler feature extraction and relates to the field of residual network, and comprises the following steps: step one: Euler feature extraction; the waveform diagram generated by EFE processing of original fault data can be directly used as the input of a Resnet50 network; step two: ResNet50 network training; the feature image processed by EFE is input into a Resnet50 network model, and the parameters thereof are trained; step three: fault diagnosis; the role of a ResNet50 classifier is played by Linear, which classifies fault types after training the network parameters in the previous training stage. Further, the ResNet50 network contains various important basic units in a CNN; meanwhile, the method can be applied to the fault diagnosis of power electronic circuits and rotating bearings. Euler feature extraction can enhance the spatial information of time series fault data, so that faults can be clearly distinguished.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of residual network, and particularly relates to a fault diagnosis method based on Euler feature extraction. BACKGROUND

[0002] The running state of core components can usually be reflected by its running data, such as the rolling bearing in mechanical equipment and the change of capacitive operating parameters in power electronic circuits. Through accurate judgment of the fault type, the characteristics of the ongoing fault can be revealed to help maintenance personnel to quickly overhaul and avoid the occurrence of major accidents. In recent years, various fault diagnosis methods have been proposed, which usually include two parts: feature extraction and fault classification. Among the commonly used feature extraction methods are empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD) and variational mode decomposition (VMD). After obtaining the modal combination of different frequencies, it is linearly recombined as a fault feature to input a support vector machine (SVM) or a back propagation neural network (BPNN) for fault classification.

[0003] However, the accuracy of the fault diagnosis method based on artificial neural network depends largely on the effectiveness of the extracted features and expert knowledge, and the diagnosis process is complex. With the rapid development of deep learning in the field of image and the like, its excellent deep feature extraction capability and processing capacity of massive data lead to the research of new methods for circuit fault and bearing fault diagnosis. SUMMARY

[0004] To solve the problems mentioned in the background, the purpose of the present application is to provide a fault diagnosis method based on Euler feature extraction, which highlights the fault features through the Euler feature extraction formula.

[0005] The purpose of the present application can be achieved by the following technical solutions:

[0006] A fault diagnosis method based on Euler feature extraction, comprising the following steps:

[0007] Step one: Euler feature extraction, the waveform diagram generated by EFE processing of the original fault data can be directly used as the input of the Resnet50 network;

[0008] Step two: training of ResNet50 network, the feature image processed by EFE is input into the Resnet50 network model, and the parameters thereof are trained.

[0009] Step three: fault diagnosis, the role of the ResNet50 classifier is played by Linear, which classifies the fault type after training the network parameters in the previous training stage.

[0010] Further, the ResNet50 network contains various important basic units in CNN.

[0011] Further, the Euler eigenvalue extraction includes the Euler eigenvalue extraction formula using data and time information based on Euler formula

[0012] e iπ = cosφ + isinφ (1)

[0013] Where i is the time point corresponding to the time series signal, and φ is the angle cosine corresponding to the time series signal.

[0014] The Euler eigenvalue extraction formula is as follows:

[0015] First, for a given time series X = {x1, x2, …, x _n}, first, compress all its values in the interval [-1, 1];

[0016]

[0017] In the formula, x i is the original time series, and x is the compressed time series.

[0018] Second, the compressed values are encoded as angle cosine.

[0019]

[0020] Finally, combined with the angle cosine, the original time series and the compressed time series, two Euler eigenvalue extraction formulas are obtained.

[0021]

[0022] Third, for FirstEuler, we deform it into three different FirstEuler, namely A-FirstEuler, B-FirstEuler and C-FirstEuler.

[0023]

[0024] Further, the fault diagnosis is verified on two completely different fault data sets, power electronic circuit fault data set and rolling bearing fault data set:

[0025] Through cross validation, three performance indicators are used to measure the diagnostic performance of GADF-2DCNN model. Through the influence of ACC, P and R three indicators, the accuracy, precision and recall of test set are represented respectively, and the classification effect of the model is evaluated comprehensively. Their definitions are as follows:

[0026]

[0027]

[0028]

[0029] Wherein TP, FP, TN, FN are true positive, false positive, true negative, false negative sample number, accuracy, recall and precision range from 0 to 1, the greater the value, the better the fault diagnosis performance.

[0030] The beneficial effects of the application are:

[0031] The application can be applied to the fault diagnosis of power electronic circuits and rotating bearings. Euler feature extraction can enhance the spatial information of time series fault data, so that the faults can be clearly distinguished.

[0032] The fault diagnosis method based on Euler feature extraction and residual network has higher fault diagnosis accuracy than other typical feature extraction-based fault diagnosis methods, and at least increases by 3.56% and 10.49% respectively on two data sets.

[0033] Due to the excellent performance of GADF on fault data features, we use GADF to process circuit data and bearing data, and then use Resnet50 to classify faults and compare with the Euler feature extraction method. The results show that the Euler feature extraction method has good extraction effect on circuit fault and rotating bearing fault. Compared with the diagnosis model using GADF, the diagnosis accuracy of the model on two data sets is increased by 3.95% and 0.56% respectively. BRIEF DESCRIPTION OF DRAWINGS

[0034] The application will be further described below in conjunction with the drawings.

[0035] Figure 1 is a model overall flow chart;

[0036] Figure 2 is a Resnet50 network structure diagram;

[0037] Figure 3 is the confusion matrix of the A-FirstEuler circuit fault diagnosis experimental results;

[0038] Figure 4 is the confusion matrix used for A-FirstEuler bearing fault diagnosis experimental results;

[0039] Figure 5 is when stride=500, the confusion matrix of the experimental results of SecondEuler used for circuit fault diagnosis;

[0040] Figure 6 is the confusion matrix of the diagnosis results of the bearing fault SecondEuler when = 3;

[0041] Figure 7 is the comparison of the experimental results of GADF in the bearing fault diagnosis;

[0042] Figure 8 is the comparison of the experimental results of GADF in the circuit fault diagnosis; DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0044] In the description of the present application, it should be understood that the terms "opening", "upper", "lower", "thickness", "top", "middle", "length", "inner", "periphery" and the like indicate the orientation or positional relationship, which are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the indicated component or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0045] A fault diagnosis method based on Euler feature extraction, comprising the following steps:

[0046] Step one: Euler feature extraction, the waveform diagram generated by EFE processing of the original fault data can be directly used as the input of the Resnet50 network;

[0047] Step two: training of the ResNet50 network, the feature image processed by EFE is input into the Resnet50 network model, and the parameters thereof are trained.

[0048] Step three: fault diagnosis, the role of the ResNet50 classifier is played by Linear, which classifies the fault type after training the network parameters in the previous training stage.

[0049] The ResNet50 network contains various important basic units in CNN.

[0050] Euler eigenvalue extraction includes Euler eigenvalue extraction formula using data and time information based on Euler formula

[0051] e iπ = cosφ + isinφ (1)

[0052] where i is the time point corresponding to the time series signal, and φ is the angle cosine corresponding to the time series signal.

[0053] The Euler feature extraction formula is as follows:

[0054] First, for a given time series X = {x1, x2, …, x _n}, first, compress all its values in the interval [-1, 1];

[0055]

[0056] where x i is the original time series, and x is the compressed time series.

[0057] Second, the compressed values are encoded as angle cosine.

[0058]

[0059] Finally, combined with the angle cosine, the original time series and the compressed time series, two Euler feature extraction formulas are obtained.

[0060]

[0061] Third, for FirstEuler, we deform it into three different FirstEuler, namely A-FirstEuler, B-FirstEuler and C-FirstEuler.

[0062]

[0063] Fault diagnosis is verified on two completely different fault datasets, power electronic circuit fault dataset and rolling bearing fault dataset:

[0064] Through cross-validation, three performance indicators are used to measure the diagnostic performance of GADF-2DCNN model. Through the influence of ACC, P and R three indicators, the accuracy, precision and recall rate of test set are represented, and the classification effect of the model is comprehensively evaluated. Their definitions are as follows:

[0065]

[0066]

[0067]

[0068] Where TP, FP, TN, FN are the number of true positive, false positive, true negative, false negative samples, the accuracy, recall rate and precision range from 0 to 1, the larger the value, the better the fault diagnosis performance.

[0069] To verify the performance of this method, we used a ResNet50 network to validate it on circuit fault data and bearing fault data. The circuit fault dataset contains 16 faults, the bearing fault dataset contains data at 1730 r / min and 1772 r / min, and the mixed-condition dataset contains two speeds. The mixed-condition dataset contains 18 fault data and two sets of normal data.

[0070] 1. FirstEuler Validation

[0071] (1) Circuit fault verification experiment

[0072] In Table 1, A represents A-FirstEuler, B represents B-FirstEuler, and C represents C-FirstEuler. The experimental results show that, with the same step size, FirstEuler (represented by A-FirstEuler) is detected by ResNet50 after a data step size of 500. After data processing, ResNet50 is used for fault diagnosis, and the diagnostic results are good, with a diagnostic accuracy of 98.85%. Its confusion matrix is ​​shown below. Figure 4 As shown.

[0073] Table 1. Fault Diagnosis Results of A-First Euler Circuit

[0074]

[0075] Table 2. Specific indicators for various diagnostic results of circuits in A-FirstEule

[0076]

[0077] (2) Bearing fault diagnosis experiment

[0078] From the preceding introduction, we can observe that among the three variants of FirstEuler, A-FirstEuler achieves the highest diagnostic accuracy across all steps compared to the other two. Therefore, we selected A-FirstEuler as the method for processing bearing data, where CWRU1730+CWRU1772 represents experimental data for mixed-condition fault diagnosis, containing bearing fault data at two different speeds. After classifying bearing faults under a single condition using A-FirstEuler, the ResNet50 fault classification results reached 99.44% and 100%, respectively, while the diagnostic accuracy still reached 96.39% when the two conditions were mixed. Its confusion matrix is ​​as follows: Figure 4 As shown.

[0079] Table 3. Experimental Results of A-First Euler Bearing Fault Diagnosis

[0080]

[0081] 2. SecondEuler verification

[0082] Since SecondEuler contains adjustable parameters, we choose The Euler feature extraction formula, formed at times 1, 3, 5, and 7, extracts features from fault data and generates feature maps, allowing observation of different... The diagnostic results obtained at that time were used to select the best performing SecondEuler.

[0083] (1) Circuit Fault Diagnosis Experiment

[0084] Table 4: Experimental Results of Circuit Fault Diagnosis under Different Parameters for SecondEuler

[0085]

[0086] As shown in Table 4, when When the data truncation step size is 500, SecondEuler's circuit fault diagnosis accuracy can reach 98.54%, therefore it is selected. The value is 3. Its confusion matrix is ​​as follows: Figure 5 As shown.

[0087] (2) Bearing fault diagnosis experiment

[0088] Similarly, we selected the best-performing SecondEuler from the circuit dataset for fault diagnosis experiments on a bearing fault dataset under mixed conditions, i.e., when... SecondEuler at that time. The diagnostic results are shown in the table below. In the bearing fault diagnosis experiment under both mixed operating conditions, the diagnostic accuracy reached 96.39%. The confusion matrix is ​​as follows. Figure 6 As shown.

[0089] Table 5. When Bearing fault diagnosis results using the SecondEuler method

[0090]

[0091] 3. Comparison with other methods

[0092] (1) Comparison with GADF

[0093] Gram angular difference field (GADF) can preserve the transient dependencies and temporal correlations between sequences, thereby obtaining high-dimensional features for feature extraction and approximate time series reconstruction. This is a good feature extraction method that can highlight the characteristics of the data. To further demonstrate the superior performance of our proposed Euler feature extraction formula, comparative experiments were conducted on circuit fault data and mixed bearing state data. Fault data processed by the three methods were then input into ResNet50 for fault classification. The experimental results are shown in Figure 7. Figure 8 As shown, it is clear that the classification accuracy obtained by the Euler feature extraction method is higher than that obtained by using GADF to process fault data, whether for circuit fault data or bearing fault data. This further verifies the superiority of Euler feature extraction.

[0094] (2) Comparison with existing feature extraction methods

[0095] In summary, the Euler feature extraction method is a feature extraction method, therefore we compared it with other feature extraction methods such as EMD and VMD, and conducted comparative experiments on two fault datasets using ELM as the classifier. The results clearly show that the Euler feature extraction method performs better on both datasets. The results are shown in Table 10.

[0096] Table 6 compares the experimental results with those of traditional feature extraction methods.

[0097]

[0098] (2) Comparison with existing fault diagnosis methods

[0099] To verify the advancement of the proposed fault diagnosis method based on Euler feature extraction, we conducted comparative experiments with other fault diagnosis methods on the DC-DC circuit fault dataset and the CWRU bearing fault dataset. The results are shown in Table 7. We can see that our two proposed Euler feature extraction methods achieve more accurate results on both datasets compared to the comparison methods. A-Firsteuler achieves a maximum improvement of 47.60% in diagnostic accuracy on the circuit fault dataset and a maximum improvement of 30.89% on the bearing fault dataset. This demonstrates the advanced properties of Euler feature extraction.

[0100] Table 7. Experimental results compared with traditional fault diagnosis

[0101]

[0102] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0103] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A fault diagnosis method based on Euler feature extraction, characterized in that, Includes the following steps: Step 1: Euler feature extraction. The waveform generated by EFE processing of the raw fault data can be directly used as input to the ResNet50 network. Step 2: Training the ResNet50 network. Input the feature image processed by EFE into the ResNet50 network model and train its parameters. Step 3: Fault diagnosis. The role of the ResNet50 classifier is played by the Linear classifier, which classifies the fault type after training the network parameters in the previous training stage. The Euler eigenvalue extraction includes a proposed Euler feature extraction formula based on Euler's formula, utilizing data and time information. e iπ =cosφ+isinφ (1) Where i is the time point corresponding to the time series signal, and φ is the angle cosine corresponding to the time series signal; The Euler feature extraction formula is as follows: First, for a given time series X = {x1, x2, ..., x...} _n First, compress all its values ​​into the range [-1, 1]. In the formula x i The original time series is... Compressed time series; Second, the compressed value is encoded as angle cosine; Finally, by combining the angle cosine, the original time series, and the compressed time series, two Euler feature extraction formulas are obtained. Third, for FirstEuler, we transform it into three different FirstEuler, namely A-FirstEuler, B-FirstEuler and C-FirstEuler; 2. The fault diagnosis method based on Euler feature extraction according to claim 1, characterized in that, The ResNet50 network contains various important basic units of CNN.

3. The fault diagnosis method based on Euler feature extraction according to claim 1, characterized in that, The fault diagnosis was validated on two completely different fault datasets: a power electronic circuit fault dataset and a rolling bearing fault dataset. Cross-validation was used to measure the diagnostic performance of the GADF-2DCNN model using three performance metrics. The impact of ACC, P, and R metrics represents the accuracy, precision, and recall on the test set, respectively, to comprehensively evaluate the model's classification performance. Their definitions are as follows: TP, FP, TN, and FN represent the number of true positive, false positive, true negative, and false negative samples, respectively. Accuracy, recall, and precision range from 0 to 1, with higher values ​​indicating better fault diagnosis performance.

Images