Mechanical fault migration diagnosis method based on variance difference expression index

Abstract

The present application relates to a kind of mechanical fault migration diagnosis method based on variance difference expression index, belong to mechanical fault migration diagnosis technical field.The method includes: S1: the original vibration signal on mechanical equipment is collected by sensor, then the original vibration signal collected is expanded sample using sliding window sampling technique;S2: variance difference expression index is used to construct the migration diagnosis model based on one-dimensional convolutional neural network;S3: the training sample divided is input into the migration diagnosis model constructed, and the classification loss of source domain labeled sample and the distribution alignment loss between source domain and unlabeled target domain are used to iteratively update training to the migration diagnosis model constructed;S4: after multiple iteration training, error curve tends to be stable, model training is completed, and the migration diagnosis model trained will be used for the closed set migration diagnosis of mechanical equipment.The present application can improve the accuracy and robustness of mechanical fault migration diagnosis.

Description

Technical Field

[0001] This invention belongs to the field of mechanical fault migration diagnosis technology, and relates to a mechanical fault migration diagnosis method based on variance difference expression index. Background Technology

[0002] Rotating machinery is widely used in important engineering fields such as energy and power, rail transportation, military defense, and aerospace. However, some key mechanical components, such as bearings, gears, and shafts, often operate under harsh conditions of high speed, heavy load, and instability. Failures can cause irreparable economic and property losses, and even casualties. In recent years, due to the lack of prior knowledge regarding mechanical fault labels, mechanical fault transfer diagnosis methods that consider data distribution differences have received considerable attention from experts and scholars. However, distribution difference indices, as a core factor affecting the accuracy of transfer diagnosis methods, have received relatively little research. Actual mechanical equipment typically operates in noisy and variable environments, resulting in low robustness and anti-interference capabilities of existing mainstream distribution difference indices.

[0003] Therefore, there is an urgent need for an improved index with anti-interference distribution difference to solve the problem of poor robustness in mechanical fault migration diagnosis. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a mechanical fault migration diagnosis method based on a variance discrepancy representation index, which uses an interference-resistant distribution difference index with variance discrepancy representation (VDR) to improve the robustness and accuracy of mechanical fault migration diagnosis.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A mechanical fault migration diagnosis method based on variance difference index, characterized in that the method specifically includes the following steps:

[0007] S1: Collect raw vibration signals from mechanical equipment using sensors, and then expand the sample using sliding window sampling technology.

[0008] S2: Construct a transfer diagnostic model based on a one-dimensional convolutional neural network according to the variance difference expression index;

[0009] S3: Input the segmented training samples into the constructed transfer diagnostic model, and use the classification loss L of the labeled samples in the source domain. C Distribution alignment loss L between the source domain and the unlabeled target domain VDR The constructed transfer diagnostic model is iteratively updated and trained.

[0010] S4: After multiple iterations of training, the error curve tends to stabilize, the model training is complete, and the trained transfer diagnosis model will be used for closed-set transfer diagnosis of mechanical equipment.

[0011] Furthermore, in step S2, the constructed transfer diagnostic model utilizes a one-dimensional convolutional neural network as the backbone to extract fault features, and all loss optimization terms are added to the last layer; wherein, the model selects classification loss L C To train source domain labeled samples, separable fault features are obtained; biased variance difference is used as the distribution alignment loss L. VDR To reduce the distribution differences between the source and target domains.

[0012] Furthermore, classification loss L C The expression is:

[0013]

[0014] Among them, C and Let n represent the source domain sample fault type and the predicted label of the source domain sample, respectively. S Indicates the number of samples in the source domain. I represents the label corresponding to the source domain sample, and I(·) represents the indicator function.

[0015] Furthermore, the distribution alignment loss L VDR The expression is:

[0016]

[0017] in, Representing Hilbert space, Represents the tensor product; Represents the source domain. Indicates the target domain;

[0018] The expression for biased variance is given by the following formula:

[0019]

[0020] Where τ(x,·) represents the kernel function, m represents the number of samples in sample set X, n represents the number of samples in sample set Y, and x i ,x j Let y represent the i-th and j-th samples in the sample set X, respectively. i ,y j Let X and Y represent the i-th and j-th samples in the sample set Y, respectively, and let X and Y represent the sample sets of the two domains.

[0021] Furthermore, the expression for the kernel function τ(x,·) is:

[0022]

[0023] in, d represents the degrees of freedom, Γ(·) represents the gamma function, and x and y represent samples in sample sets X and Y; E represents the tensor product. xp(x) κ(x,·) represents the expectation of κ(x,·) under the distribution p(x), and κ(x,y) represents the interference-resistant student kernel function.

[0024] Furthermore, in step S3, Adam is used to optimize the weight parameters Θ of the transfer diagnostic model:

[0025]

[0026] Where ε and λ represent the learning rate and the tradeoff parameter, respectively.

[0027] The beneficial effects of this invention are as follows: compared with the current transfer diagnosis model constructed by typical distribution difference indices, the transfer diagnosis method based on variance difference expression index (VDR) of this invention has higher transfer diagnosis accuracy and stronger generalization ability.

[0028] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0029] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0030] Figure 1 This is a schematic diagram of data mapping from two-dimensional space to three-dimensional space.

[0031] Figure 2 This is a schematic diagram of the transfer diagnosis model based on variance difference expression index of the present invention;

[0032] Figure 3 For CWRU test bench;

[0033] Figure 4 For IM test bench;

[0034] Figure 5 For DDS test bench. Detailed Implementation

[0035] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0036] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0037] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0038] Please see Figures 1-5 This invention provides a cross-bearing migration diagnosis method based on variance difference index, the specific process of which is as follows:

[0039] 1) Acquire raw vibration signals of bearings on mechanical equipment using accelerometers to facilitate subsequent migration diagnostic tasks. Then, expand the sample using sliding window sampling technology.

[0040] 2) Design a variance discrepancy representation (VDR) composed of equations (11) and (8).

[0041] Maximum mean discrepancy (MMD) is currently the mainstream distributional discrepancy index in transfer diagnostics. The term "maximum" comes from spatial mapping, specifically mapping data points from a low-dimensional sample space to an infinite-dimensional Hilbert space using a kernel function. According to pattern recognition theory, the higher the spatial dimension, the greater the separability of the data. Figure 1 The data mapping from two-dimensional space to three-dimensional space is shown below. The formula for the maximum average difference is expressed as follows:

[0042]

[0043] in, Representing Hilbert space, and Let h and h represent two data sample domains, respectively, and h(·) represent the mapping function. Given and and Let represent the sample set of the corresponding domain, and p(x) and q(y) represent the marginal probability distributions of the corresponding domain. Equation (2) can be used to obtain:

[0044]

[0045] Where κ(·,·) represents the binary kernel function of the corresponding domain, which usually refers to the Gaussian kernel function; u p and u q These represent the kernel mean embeddings of the corresponding domains.

[0046] The maximum mean difference is mainly a distributional difference index based on the evolution of the mean statistic. However, the monitoring data collected from actual mechanical equipment is usually a one-dimensional, symmetrical original vibration signal along the Y-axis, meaning the mean is always 0. Therefore, the maximum mean difference is insensitive to the variation of mechanical condition monitoring signals. Based on the above problem, this invention constructs a distributional difference index that can reflect the variation of signal variance. First, a kernel function is constructed, as shown in the following equation:

[0047]

[0048] in, This represents the tensor product. Using the kernel function described above, the final variance variance index can be obtained:

[0049]

[0050] in, To ensure the existence of the supremum on the right-hand side of the equation, the unit regenerated Hilbert space is chosen as the constraint term. Clearly, since the tensor product is a generalized operation of orders of magnitude, equation (4) will fully reflect the variance difference between the two data domains. For easier calculation, equation (4) can be rewritten as a squared generalized variance difference index:

[0051]

[0052] Considering that the sampled data is finite in practice, this invention provides a description of biased variance. and unbiased variance variability statistics

[0053]

[0054]

[0055] Equations (3), (6), and (7) show that the final distribution difference index's difference measurement effect largely depends on the base kernel function κ(x,·), and the commonly used base kernel function is the Gaussian kernel function derived from the Gaussian distribution. When there are outliers deviating from the sample center, the Gaussian kernel function will deviate from the main data, i.e., its robustness is poor. Therefore, this invention constructs an interference-resistant student kernel function based on the long-tailed distribution of the student distribution:

[0056]

[0057] Where d represents the degrees of freedom, and its range is d > 0; Γ(·) represents the gamma function:

[0058]

[0059] 3) Construct a transfer diagnostic model based on a one-dimensional convolutional neural network, using the variance difference index developed in the previous step, such as... Figure 2 As shown.

[0060] 4) Input the segmented training samples into the constructed transfer diagnostic model, and use the classification loss L of the labeled samples in the source domain. C (Equation 10) Distribution alignment loss L between the source domain and the unlabeled target domain VDR (Equation 11) The constructed transfer diagnosis model is iteratively updated and trained (Equation 12).

[0061] like Figure 2As shown, the transfer diagnostic model based on the variance difference index uses a one-dimensional convolutional neural network as the backbone to extract fault features. To save computational resources, all loss optimization terms are added to the last layer. To fully test the effectiveness of the variance difference index, it is assumed that the source domain is labeled. The target domain is unlabeled. and n S and n T These represent the number of samples and the number of samples in the corresponding domain, respectively. This represents the label corresponding to the source domain sample. This invention selects classification loss L... C Separable fault features are obtained by training source domain labeled samples.

[0062]

[0063] Among them, C and Let I(·) represent the source domain sample fault type and the predicted label of the source domain sample, respectively, and let I(·) represent the indicator function. Secondly, this invention uses biased variance difference as the distribution alignment loss L. VDR To reduce the distribution differences between the source and target domains:

[0064]

[0065] Then use Adam to optimize. Figure 2 The weight parameters Θ of a one-dimensional convolutional neural network:

[0066]

[0067] Where ε and λ represent the learning rate and the tradeoff parameter, respectively.

[0068] 5) After multiple iterations of training, the error curve tends to stabilize, the model training is completed, and the trained transfer diagnosis model will be used for closed-set transfer diagnosis across bearings.

[0069] The above is the flowchart of the cross-bearing migration diagnostic model based on variance difference expression index proposed in this invention. The experimental results below have demonstrated the effectiveness of this intelligent diagnostic method.

[0070] Verification Experiment: The raw bearing vibration signals collected in this experiment came from three test benches, specifically the CWRU standard bearing dataset publicly available from Case Western Reserve University. A schematic diagram of the CWRU bearing dataset is shown below. Figure 3 As shown, it consists of a motor, bearings at both ends of the motor, a torque sensor, and a power meter;

[0071] The IM bearing dataset comes from Jiangnan University. A schematic diagram of the IM test bench is shown below. Figure 4As shown, it consists of a servo motor, a coupling, a rotor, and bearings and bearing housings at both ends.

[0072] And the DDS bearing dataset from Southeast University, with a schematic diagram of the test bench as shown below. Figure 5 As shown, the DDS test bench mainly consists of five parts: a motor, a planetary gearbox, a parallel gearbox, and a magnetic powder brake. Different operating condition signals can be simulated by applying load through the magnetic powder brake.

[0073] Table 1 shows the fault type, fault size, and sampling frequency for each bearing dataset, where NC, IF, BF, and OF represent normal condition, inner ring fault, rolling element fault, and outer ring fault, respectively. Using these three bearing datasets, we can establish six cross-bearing closed-set transfer diagnostic tasks (A→B, B→A, A→C, C→A, C→B, and B→C).

[0074] Table 1. Detailed information for the three bearing datasets.

[0075]

[0076] Comparative experiment:

[0077] To demonstrate the superiority of the Variance Difference Representation Index (VDR)-based transfer diagnostic method of this invention, experimental results for six cross-bearing closed-set transfer diagnostic tasks are shown in Table 2, compared with current typical distribution difference indices. Table 2 shows that the Variance Difference Representation Index proposed in this invention has higher transfer diagnostic accuracy and stronger generalization ability (bold indicates the highest diagnostic accuracy in each transfer diagnostic task). Specifically, the average diagnostic accuracy of VDR on the six cross-bearing closed-set transfer diagnostic tasks reached over 87%, representing an overall improvement of approximately 45% in diagnostic accuracy compared to other transfer diagnostic methods.

[0078] Table 2 Experimental Results

[0079]

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for diagnosing mechanical fault migration based on variance difference index, characterized in that, The method specifically includes the following steps: S1: Collect raw vibration signals from mechanical equipment using sensors, and then expand the sample using sliding window sampling technology. S2: Construct a transfer diagnostic model based on a one-dimensional convolutional neural network according to the variance difference expression index; S3: Input the segmented training samples into the constructed transfer diagnostic model, and use the classification loss of the labeled samples in the source domain. Distribution alignment loss between source domain and unlabeled target domain The constructed transfer diagnostic model is iteratively updated and trained. Classification loss The expression is: in, and These represent the fault type of the source domain sample and the predicted label of the source domain sample, respectively. Indicates the number of samples in the source domain. This represents the label corresponding to the source domain sample. Indicates an indicator function; Distribution Alignment Loss The expression is: in, , Representing Hilbert space, " represents the tensor product; Represents the source domain. Indicates the target domain; The expression for biased variance is given by the following formula: in, Represents the kernel function. m Represents the sample set X The number of samples, n Represents the sample set Y The number of samples, Representing sample sets respectively X The first in i The and the first j One sample, Representing sample sets respectively Y The first in i The and the first j One sample, X , Y These represent sample sets from the two domains, respectively. S4: After multiple iterations of training, the error curve tends to stabilize, the model training is complete, and the trained transfer diagnosis model will be used for closed-set transfer diagnosis of mechanical equipment.

2. The mechanical fault migration diagnosis method according to claim 1 is characterized in that, In step S2, the constructed transfer diagnostic model uses a one-dimensional convolutional neural network as the backbone to extract fault features, and all loss optimization terms are added to the last layer; among them, the model selects classification loss. To train source domain labeled samples, separable fault features are obtained; biased variance difference is used as the distribution alignment loss. To reduce the distribution differences between the source and target domains.

3. The mechanical fault migration diagnosis method according to claim 1 is characterized in that, Kernel function The expression is: in, Indicates degrees of freedom. Represents the gamma function. x , y Represents the sample set X and Y The sample in; " represents the tensor product, Indicates to In distribution Seeking expectations This represents the student kernel function, which is resistant to interference.

4. The mechanical fault migration diagnosis method according to claim 1 is characterized in that, In step S3, Adam is used to optimize the weight parameters of the transfer diagnostic model. : in, and These represent the learning rate and the tradeoff parameter, respectively.

Images