Fault diagnosis method for electromechanical equipment

Abstract

The invention discloses a fault diagnosis method for electromechanical equipment, and belongs to the technical field of fault diagnosis, and the method comprises the following steps: S1, obtaining electromechanical equipment sensing data of a plurality of fault types to be recognized; s2, performing FFT (Fast Fourier Transform) on the sensing data of each electromechanical device in the S1, and encoding the sensing data into feature vectors; s3, constructing a KNN graph with an edge weight for the plurality of feature vectors obtained in the S2, and obtaining a KNN graph to be recognized; and S4, inputting the KNN graph to be identified into the trained DGAT model and the full connection layer, and obtaining the fault type of the sensing data of each electromechanical device in the S1. According to the method, high-accuracy fault diagnosis is realized by using a small part of labeled samples.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and more particularly to a fault diagnosis method for electromechanical equipment. Background Art

[0002] Currently, existing electromechanical equipment fault diagnosis models rely heavily on labeled samples during training, often requiring extensive domain prior knowledge and auxiliary information to construct effective diagnostic systems. While these methods have achieved some success in fault diagnosis, their overreliance on specialized knowledge not only complicates model structures, increasing the difficulty of development and maintenance, but also introduces greater uncertainty, compromising the stability and reliability of diagnostic results.

[0003] Therefore, how to provide a fault diagnosis method for electromechanical equipment that can achieve high-accuracy fault diagnosis using a small number of labeled samples is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of this, an object of the present invention is to provide a fault diagnosis method for electromechanical equipment.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In one aspect, a method for diagnosing a fault in an electromechanical device is provided, comprising the following steps:

[0007] S1: Acquire sensor data of several electromechanical devices of the type of fault to be identified;

[0008] S2: Perform FFT transformation on each electromechanical device sensor data in S1 and encode it into a feature vector;

[0009] S3: construct a KNN graph with edge weights for the several feature vectors obtained in S2 to obtain the KNN graph to be identified;

[0010] S4: Input the KNN graph to be identified into the trained DGAT model and the fully connected layer to obtain the fault type of the sensor data of each electromechanical device in S1.

[0011] Preferably, the trained DGAT model is obtained based on the following steps:

[0012] Based on a number of unlabeled samples, a first KNN graph is obtained; wherein each unlabeled sample only includes sample features; the sample features are feature vectors encoded by the sensor data of the electromechanical device after FFT transformation;

[0013] Based on the first KNN graph, the first GAT model, the second GAT model and the fully connected layer, obtaining pseudo labels for each unlabeled sample and updated sample features for each unlabeled sample;

[0014] Based on all samples with pseudo labels and updated sample features, a second KNN graph is obtained;

[0015] Obtaining a third KNN graph based on the second KNN graph and the LIME model;

[0016] Obtaining a fourth KNN graph based on the edge weights in the third KNN graph;

[0017] Based on several labeled samples, a fifth KNN graph is obtained; each labeled sample includes a sample feature and a label; the sample feature is a feature vector encoded by the electromechanical equipment sensor data after FFT transformation; the label is the fault type; the number of labeled samples is less than the number of unlabeled samples;

[0018] The DGAT model is trained based on the fourth KNN graph and the fifth KNN graph to obtain the trained DGAT model.

[0019] Preferably, an unbounded weighted KNN graph is constructed for a number of unlabeled samples to obtain a first KNN graph; wherein each unlabeled sample only includes sample features; the sample features are feature vectors encoded after the electromechanical equipment sensor data is transformed by FFT.

[0020] Preferably, the first KNN graph is sequentially input into the first GAT model, the second GAT model and the fully connected layer to obtain pseudo labels of each unlabeled sample and updated sample features of each unlabeled sample.

[0021] Preferably, a KNN graph with edge weights is constructed for all samples with pseudo labels and updated sample features to obtain a second KNN graph.

[0022] Preferably, the second KNN graph is input into a LIME model to screen out samples whose contribution is greater than a preset threshold, thereby obtaining a third KNN graph.

[0023] Preferably, L1 regularization is performed on the edge weights in the third KNN graph to obtain a fourth KNN graph with updated edge weights.

[0024] Preferably, a KNN graph with edge weights is constructed for several labeled samples to obtain a fifth KNN graph; wherein each labeled sample includes a sample feature and a label; the sample feature is a feature vector encoded after the electromechanical equipment sensor data is transformed by FFT; the label is the fault type; the number of labeled samples is less than the number of unlabeled samples.

[0025] Preferably, the fourth KNN graph and the fifth KNN graph are input into a DGAT model and a fully connected layer, and the DGAT model is trained to obtain the trained DGAT model.

[0026] On the other hand, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for diagnosing faults in electromechanical equipment when executing the computer program.

[0027] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for fault diagnosis of electromechanical equipment, which achieves high-accuracy fault diagnosis using only a small number of labeled samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0029] Figure 1 A flowchart of a method for diagnosing a fault in an electromechanical device provided by an embodiment of the present invention;

[0030] Figure 2 A flowchart of a DGAT model for training provided in an embodiment of the present invention;

[0031] Figure 3 A schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0033] On the one hand, if Figure 1 As shown, the embodiment of the present invention discloses a fault diagnosis method for electromechanical equipment, comprising the following steps:

[0034] S1: Acquire sensor data of several electromechanical devices of the type of fault to be identified;

[0035] In one embodiment, the electromechanical device is a planetary gearbox, and the sensing data of the electromechanical device is a vibration signal of the planetary gearbox collected by a sensor.

[0036] S2: Perform FFT transformation on each electromechanical device sensor data in S1 and encode it into a feature vector;

[0037] Specifically: First, the time domain electromechanical equipment sensing data is converted into frequency domain features using FFT transformation, and then the frequency domain features are encoded into feature vectors.

[0038] S3: construct a KNN graph with edge weights for the several feature vectors obtained in S2 to obtain the KNN graph to be identified;

[0039] It can be understood that: a feature vector corresponds to a node in the KNN graph;

[0040] Specifically: the KNN graph to be identified is obtained based on the following steps:

[0041] Calculate the Euclidean distance between node i and node j (the remaining nodes except node i), take the nodes corresponding to the K smallest distances among the j Euclidean distances as the K neighboring nodes of node i, establish the connection relationship between node i and the K neighboring nodes, and obtain the unbounded weighted KNN graph:

[0042] Based on the formula Calculate the edge weight between node i and node j to obtain the edge-weighted KNN graph (i.e., the KNN graph to be identified); where e ij represents the weight of the edge connecting node i and node j (i.e. edge weight); x i represents the feature vector corresponding to node i; x j represents the eigenvector corresponding to node j; ||x i -x j || represents x i -x j The Euclidean distance of node i is σ, which is the bandwidth parameter of the Gaussian kernel. N(i) represents the set of neighboring nodes of node i.

[0043] S4: Input the KNN graph to be identified into the trained DGAT model and the fully connected layer to obtain the fault type of the sensor data of each electromechanical device in S1.

[0044] In one embodiment, if Figure 2 As shown, the trained DGAT model is obtained based on the following steps:

[0045] Based on a number of unlabeled samples, a first KNN graph is obtained; wherein each unlabeled sample only includes sample features; the sample features are feature vectors encoded by the sensor data of the electromechanical device after FFT transformation;

[0046] In one embodiment, an unweighted KNN graph is constructed for a number of unlabeled samples to obtain a first KNN graph; wherein each unlabeled sample only includes sample features; the sample features are feature vectors encoded after the electromechanical device sensor data is transformed by FFT.

[0047] In one embodiment, the electromechanical device is a planetary gearbox, and the sensing data of the electromechanical device is a vibration signal of the planetary gearbox collected by a sensor.

[0048] In one embodiment, the time-domain electromechanical device sensing data is first converted into frequency-domain features using FFT transformation, and then the frequency-domain features are encoded into feature vectors.

[0049] It can be understood that: an unlabeled sample corresponds to a node in the KNN graph;

[0050] Specifically: The first KNN graph is obtained based on the following steps:

[0051] Calculate the Euclidean distance between node m and node n (the remaining nodes except node n), take the nodes corresponding to the K smallest distances among the n Euclidean distances as the K neighboring nodes of node m, establish the connection relationship between node m and the K neighboring nodes, and obtain the unbounded weighted KNN graph (i.e., the first KNN graph):

[0052] Based on the first KNN graph, the first GAT model, the second GAT model and the fully connected layer, obtaining pseudo labels for each unlabeled sample and updated sample features for each unlabeled sample;

[0053] In one embodiment, the first KNN graph is sequentially input into a first GAT model, a second GAT model, and a fully connected layer to obtain pseudo labels of each unlabeled sample and updated sample features of each unlabeled sample.

[0054] Specifically, the updated sample features of each unlabeled sample are obtained based on the following formula:

[0055]

[0056] Where, e mn represents the attention coefficient between node m and node n; F represents the attention function; W represents the preset weight matrix; x m and x n represents the sample features corresponding to node m and node n respectively; N(m) represents the set of neighboring nodes of node m; α mn represents the normalized attention coefficient between node m and node n; e mk represents the attention coefficient between node m and node k; N(m)∪m represents the union of N(m) and node m; x′ mrepresents the sample features after the first update of node m; σ1 represents the activation function; x″ m represents the sample feature after the second update of node m, which is the sample feature after the unlabeled sample is updated; W' represents the preset weight matrix; x' n Represents the sample features after the first update of node n.

[0057] x″ m Input into the fully connected layer to obtain the pseudo label corresponding to the unlabeled sample m (i.e. node m), and the pseudo label is the predicted fault type.

[0058] Based on all samples with pseudo labels and updated sample features, a second KNN graph is obtained;

[0059] In one embodiment, a KNN graph with edge weights is constructed for all samples with pseudo labels and updated sample features to obtain a second KNN graph.

[0060] It can be understood that the construction method of the second KNN graph is consistent with the construction method of the KNN graph to be identified, except that the sample features updated in S42 are used to calculate the edge weights, and each sample in the second KNN graph has a pseudo label.

[0061] Obtaining a third KNN graph based on the second KNN graph and the LIME model;

[0062] In one embodiment, the second KNN graph is input into a LIME model to filter out samples whose contribution is greater than a preset threshold, thereby obtaining a third KNN graph.

[0063] Obtaining a fourth KNN graph based on the edge weights in the third KNN graph;

[0064] In one embodiment, L1 regularization is performed on the edge weights in the third KNN graph to obtain a fourth KNN graph with updated edge weights.

[0065] Based on several labeled samples, a fifth KNN graph is obtained; each labeled sample includes a sample feature and a label; the sample feature is a feature vector encoded by the electromechanical equipment sensor data after FFT transformation; the label is the fault type; the number of labeled samples is less than the number of unlabeled samples;

[0066] In one embodiment, a weighted KNN graph is constructed for several labeled samples to obtain a fifth KNN graph; each labeled sample includes a sample feature and a label; the sample feature is a feature vector encoded after the electromechanical equipment sensor data is transformed by FFT; the label is the fault type; and the number of labeled samples is less than the number of unlabeled samples.

[0067] It can be understood that the method for constructing the fifth KNN graph is consistent with the method for constructing the KNN graph to be identified, except that each sample in the fifth KNN graph has a true label.

[0068] The DGAT model is trained based on the fourth KNN graph and the fifth KNN graph to obtain the trained DGAT model.

[0069] In one embodiment, the fourth KNN graph and the fifth KNN graph are input into a DGAT model and a fully connected layer, and the DGAT model is trained to obtain the trained DGAT model.

[0070] It should be noted that the present invention compared different label ratios and found that the diagnostic accuracy of the fault type gradually increased from 3% to 11%, and the accuracy was close to 98% at 11%, where the label ratio refers to the ratio of the number of labeled samples to the number of unlabeled samples.

[0071] The present invention uses two sample sets with a label ratio of 7% to compare different K values ​​(i.e., the number of neighboring nodes selected when constructing the KNN graph) and finds that when K=5, the diagnostic accuracy of the fault type can reach 94.6% and 97.8%.

[0072] Therefore, the method of the present invention can significantly improve the diagnostic accuracy of fault types.

[0073] On the other hand, the present invention also provides an electronic device, such as Figure 3 As shown, the electronic device may include: a processor 301, a communications interface 302, a memory 303, and a communication bus 304. The processor 301, the communications interface 302, and the memory 303 communicate with each other via the communication bus 304. The processor 301 may call logic instructions in the memory 303 to execute the fault diagnosis method for the electromechanical device.

[0074] In addition, the logic instructions in the above-mentioned memory 303 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0075] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0076] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A fault diagnosis method for electromechanical equipment, characterized in that: The following steps are involved: S1: Acquire sensor data of several electromechanical devices of the type of fault to be identified; S2: Perform FFT transformation on each electromechanical device sensor data in S1 and encode it into a feature vector; S3: construct a KNN graph with edge weights for the several feature vectors obtained in S2 to obtain the KNN graph to be identified; S4: Input the KNN graph to be identified into the trained DGAT model and the fully connected layer to obtain the fault type of the sensor data of each electromechanical device in S1.

2. A fault diagnosis method for electromechanical equipment according to claim 1, characterized in that: The trained DGAT model is obtained based on the following steps: Based on a number of unlabeled samples, a first KNN graph is obtained; wherein each unlabeled sample only includes sample features; the sample features are feature vectors encoded by the sensor data of the electromechanical device after FFT transformation; Based on the first KNN graph, the first GAT model, the second GAT model and the fully connected layer, obtaining pseudo labels for each unlabeled sample and updated sample features for each unlabeled sample; Based on all samples with pseudo labels and updated sample features, a second KNN graph is obtained; Obtaining a third KNN graph based on the second KNN graph and the LIME model; Obtaining a fourth KNN graph based on the edge weights in the third KNN graph; Based on several labeled samples, a fifth KNN graph is obtained; each labeled sample includes a sample feature and a label; the sample feature is a feature vector encoded by the electromechanical equipment sensor data after FFT transformation; the label is the fault type; the number of labeled samples is less than the number of unlabeled samples; The DGAT model is trained based on the fourth KNN graph and the fifth KNN graph to obtain the trained DGAT model.

3. A fault diagnosis method for electromechanical equipment according to claim 2, characterized in that: A KNN graph without edge weights is constructed for a number of unlabeled samples to obtain a first KNN graph, wherein each unlabeled sample only includes sample features; the sample features are feature vectors encoded by the sensor data of the electromechanical equipment after FFT transformation.

4. A fault diagnosis method for electromechanical equipment according to claim 2, characterized in that: The first KNN graph is sequentially input into the first GAT model, the second GAT model and the fully connected layer to obtain pseudo labels of each unlabeled sample and updated sample features of each unlabeled sample.

5. A fault diagnosis method for electromechanical equipment according to claim 2, characterized in that: A weighted KNN graph is constructed for all samples with pseudo labels and updated sample features to obtain a second KNN graph.

6. A fault diagnosis method for electromechanical equipment according to claim 2, characterized in that: The second KNN graph is input into the LIME model to filter out samples whose contribution is greater than a preset threshold, and a third KNN graph is obtained.

7. A fault diagnosis method for electromechanical equipment according to claim 2, characterized in that: L1 regularization is performed on the edge weights in the third KNN graph to obtain a fourth KNN graph with updated edge weights.

8. A fault diagnosis method for electromechanical equipment according to claim 2, characterized in that: A weighted KNN graph is constructed for several labeled samples to obtain a fifth KNN graph. Each labeled sample includes a sample feature and a label. The sample feature is a feature vector encoded by the sensor data of the electromechanical equipment after FFT transformation. The label is the fault type. The number of labeled samples is less than the number of unlabeled samples.

9. A fault diagnosis method for electromechanical equipment according to claim 2, characterized in that: The fourth KNN graph and the fifth KNN graph are input into a DGAT model and a fully connected layer, and the DGAT model is trained to obtain the trained DGAT model.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the fault diagnosis method for an electromechanical device according to any one of claims 1 to 9 is implemented.

Images