Gearbox input shaft bearing life prediction method, device and equipment

Abstract

The invention discloses a gearbox input shaft bearing life prediction method, device and equipment, and relates to the technical field of gearboxes, and the method comprises the steps: converting a segmented gearbox input shaft bearing vibration signal into a vibration image, and extracting a sample vibration image; determining a physical model life index based on a life prediction physical model, and labeling the preset number of sample vibration images to obtain labeled vibration images and unlabeled vibration images; performing feature extraction on the labeled vibration image and the unlabeled vibration image to obtain a feature representation vector; calculating a comprehensive evaluation index of the feature representation vectors, and when the comprehensive evaluation index is greater than or equal to an evaluation index threshold, fusing the feature representation vectors to obtain fused features; and based on the fusion features, determining a life prediction result of the gearbox input shaft bearing. By means of the mode, the advantages of deep learning and the physical model are combined, the service life of the gearbox input shaft bearing is predicted, and the prediction accuracy and reliability are improved.

Description

Technical Field

[0001] The present application relates to the technical field of gearboxes, and in particular to a method, device and equipment for predicting the life of a gearbox input shaft bearing. Background Art

[0002] The gearbox input shaft bearing is one of the key components of the automobile transmission system, and its performance directly affects the operating efficiency and reliability of the vehicle. Early failure of the bearing often leads to serious mechanical damage and even safety accidents. Therefore, it is of great practical significance to predict and monitor the life of the gearbox input shaft bearing.

[0003] At present, for the life prediction of the gearbox input shaft bearing, the traditional empirical formula method has limited accuracy, is difficult to adapt to the complex and changeable actual working conditions, is overly dependent on historical data, and is not accurate enough for the bearing life prediction of new engines or special working conditions; the traditional finite element analysis method uses engineering software to perform mechanical analysis to predict the stress and strain distribution of the gearbox input shaft bearing under specific working conditions, including model establishment, boundary setting, model debugging and operation calculation, which requires a large amount of simulation resources and has a long cycle and high calculation cost, and is not suitable for real-time monitoring and rapid prediction; traditional machine learning methods are highly dependent on feature engineering, and it is difficult to process high-dimensional data and capture complex patterns in time series, and the accuracy needs to be improved; traditional deep learning methods use deep learning models such as convolutional neural networks (CNN) and recurrent neural networks (RNN) to automatically extract complex features of vibration signals for life prediction, but most of them rely on a large amount of labeled data for training. In practical applications, it is often difficult to obtain a large amount of labeled data.

[0004] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of the present application is to provide a method, device and equipment for predicting the life of a gearbox input shaft bearing, aiming to solve the technical problem of low accuracy in predicting the life of a gearbox input shaft bearing in the prior art.

[0006] To achieve the above objectives, the present application provides a method for predicting the life of a gearbox input shaft bearing, the method comprising:

[0007] Acquire vibration signals of the gearbox input shaft bearing at different times during its entire life cycle, convert the segmented vibration signals into corresponding vibration images, and extract sample vibration images from the vibration images;

[0008] Based on the life prediction physical model of the gearbox input shaft bearing, determine the physical model life index of the gearbox input shaft bearing;

[0009] Based on the physical model life index, a preset number of sample vibration images are annotated to obtain annotated vibration images and unannotated vibration images;

[0010] The annotated vibration image and the unannotated vibration image are input into a dual-branch network for feature extraction, and the corresponding feature representation vectors are obtained respectively;

[0011] Calculate the evaluation index of the feature representation vector, and determine the comprehensive evaluation index based on the evaluation index, the evaluation index at least includes feature consistency index, separability index, robustness index, overall feature consistency index and fusion quality index;

[0012] When the comprehensive evaluation index is greater than or equal to the evaluation index threshold, the feature representation vector is fused to obtain the fused feature;

[0013] Based on the fused features, the life prediction results of the gearbox input shaft bearing are determined.

[0014] In one embodiment, based on the life prediction physical model of the gearbox input shaft bearing, before the step of determining the physical model life index of the gearbox input shaft bearing, the step further includes:

[0015] Obtaining performance parameters of the gearbox input shaft bearing, the performance parameters at least including rated dynamic load, actual dynamic load, life index, engine speed, reliability correction factor and life correction factor;

[0016] Based on the first correspondence between the performance parameters and the life indicators of the physical model, a life prediction physical model for the gearbox input shaft bearing is constructed.

[0017] In one embodiment, the feature representation vector includes a first feature representation vector and a second feature representation vector, the dual-branch network includes a first branch network and a second branch network, the first branch network and the second branch network both include a coding layer, a pooling layer, and a projection head connected in sequence, and the steps of inputting the labeled vibration image and the unlabeled vibration image into the dual-branch network for feature extraction, and obtaining the corresponding feature representation vectors respectively include:

[0018] Input the dimension-reduced unlabeled vibration image into the encoding layer of the first branch network to perform feature extraction to obtain a first initial feature, and input the first initial feature into the pooling layer and the projection head of the first branch network to obtain a first feature representation vector of the unlabeled vibration image;

[0019] The annotated vibration image is input into the encoding layer of the second branch network for feature extraction to obtain a second initial feature, and the second initial feature is input into the pooling layer and the projection head of the second branch network to obtain a second feature representation vector of the annotated vibration image.

[0020] In one embodiment, based on the evaluation index, the step of determining the comprehensive evaluation index includes:

[0021] Calculate the index score of the first branch network based on the feature consistency index, the separability index, the robustness index, the weight of the feature consistency index, the weight of the separability index, and the weight of the robustness index corresponding to the first feature representation vector;

[0022] Calculate the index score of the second branch network based on the feature consistency index, the separability index, the robustness index, the weight of the feature consistency index, the weight of the separability index, and the weight of the robustness index corresponding to the second feature representation vector;

[0023] Obtaining a second corresponding relationship between the overall feature consistency index, the fusion quality index, the weight of the overall feature consistency index, the weight of the fusion quality index, the index score, the balance coefficient and the comprehensive evaluation index;

[0024] Based on the overall feature consistency index, the fusion quality index, the weight of the overall feature consistency index, the weight of the fusion quality index, the index score of the first branch network, the index score of the second branch network, the balance coefficient and the second corresponding relationship, a comprehensive evaluation index is determined.

[0025] In one embodiment, the method further comprises:

[0026] Obtaining a third correspondence between the feature representation vector, the feature mean, and the overall feature consistency index;

[0027] Determining an overall feature consistency index based on the first feature characterization vector, the feature mean of the first feature characterization vector, the second feature characterization vector, the feature mean of the second feature characterization vector, and the third corresponding relationship;

[0028] Obtaining a fourth corresponding relationship between the feature representation vector, the feature dimension number and the fusion quality index;

[0029] A fusion quality indicator is determined based on the first feature characterization vector, the second feature characterization vector, the number of feature dimensions, and the fourth corresponding relationship.

[0030] In one embodiment, the first branch network further includes a prediction layer connected to the projection head, and the method further includes:

[0031] When the comprehensive evaluation index is less than the evaluation index threshold, the first feature representation vector is input into the prediction layer of the first branch network to obtain first prediction data;

[0032] Based on the first predicted data, symmetrically determine second predicted data corresponding to the second feature characterization vector;

[0033] Calculating a first correlation coefficient between the second prediction data and the first feature characterization vector, and calculating a second correlation coefficient between the first prediction data and the second feature characterization vector based on a gradient stopping strategy of the second branch network;

[0034] Determine a contrast loss function of the dual-branch network based on the first correlation coefficient, the second correlation coefficient, and the correlation factor;

[0035] Based on the contrast loss function of the dual-branch network and the gradient stopping strategy of the second branch network, the parameters of the dual-branch network are updated, and the labeled vibration image and the unlabeled vibration image are input into the dual-branch network for feature extraction to obtain the corresponding feature representation vectors.

[0036] In one embodiment, the step of fusing the feature representation vectors to obtain the fused features includes:

[0037] Perform residual operation on the feature representation vector to obtain residual features;

[0038] The weighted residual features are connected in multiple levels to obtain the initial fusion features;

[0039] The initial fusion features are input into the feature fusion classification module to obtain the fusion features. The feature fusion classification module includes a first convolution layer, a normalization layer, a first activation layer, a channel attention layer, a pooling layer, a second convolution layer and a second activation layer connected in sequence.

[0040] In one embodiment, after the step of determining the life prediction result of the gearbox input shaft bearing based on the fusion feature, the step further includes:

[0041] Determine maintenance recommendations based on life prediction results for transmission input shaft bearings.

[0042] In addition, to achieve the above purpose, the present application also proposes a gearbox input shaft bearing life prediction device, the gearbox input shaft bearing life prediction device comprising:

[0043] A feature extraction module is used to obtain vibration signals of the gearbox input shaft bearing at different times during the entire life cycle, convert the segmented vibration signals into corresponding vibration images, and extract sample vibration images from the vibration images;

[0044] The feature extraction module is also used to determine the physical model life index of the gearbox input shaft bearing based on the life prediction physical model of the gearbox input shaft bearing;

[0045] The feature extraction module is also used to annotate a preset number of sample vibration images based on the physical model life index to obtain annotated vibration images and unannotated vibration images;

[0046] The feature extraction module is also used to input the labeled vibration image and the unlabeled vibration image into the double-branch network for feature extraction, and obtain corresponding feature representation vectors respectively;

[0047] A feature fusion module is used to calculate the evaluation index of the feature representation vector, and determine the comprehensive evaluation index based on the evaluation index. The evaluation index includes at least a feature consistency index, a separability index, a robustness index, an overall feature consistency index and a fusion quality index;

[0048] The feature fusion module is also used to fuse the feature representation vectors to obtain fused features when the comprehensive evaluation index is greater than or equal to the evaluation index threshold;

[0049] The life prediction module is used to determine the life prediction result of the gearbox input shaft bearing based on the fusion features.

[0050] In addition, to achieve the above-mentioned purpose, the present application also proposes a gearbox input shaft bearing life prediction device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the gearbox input shaft bearing life prediction method as described above.

[0051] In addition, to achieve the above-mentioned purpose, the present invention also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the gearbox input shaft bearing life prediction method as described above are implemented.

[0052] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the gearbox input shaft bearing life prediction method as described above are implemented.

[0053] The present application provides a method for predicting the life of a gearbox input shaft bearing, which includes obtaining vibration signals at different times during the entire life cycle of the gearbox input shaft bearing, converting the segmented vibration signals into corresponding vibration images, and extracting sample vibration images from the vibration images; determining the physical model life index of the gearbox input shaft bearing based on the life prediction physical model of the gearbox input shaft bearing; annotating a preset number of sample vibration images based on the physical model life index to obtain annotated vibration images and unlabeled vibration images; inputting the annotated vibration images and the unlabeled vibration images into a double-branch network for feature extraction to obtain corresponding feature representation vectors respectively; calculating evaluation indicators of the feature representation vectors, and determining a comprehensive evaluation indicator based on the evaluation indicator, the evaluation indicator at least including a feature consistency indicator, a separability indicator, a robustness indicator, an overall feature consistency indicator and a fusion quality indicator; when the comprehensive evaluation indicator is greater than or equal to an evaluation indicator threshold, fusing the feature representation vectors to obtain a fusion feature; and determining a life prediction result of the gearbox input shaft bearing based on the fusion feature. This application converts the vibration signal into a vibration image, and then extracts some vibration images as sample data for preliminary life prediction. A small number of vibration images labeled with physical model life indicators and a large number of unlabeled vibration images are compared and studied, and the multimodal heterogeneous features of the bearing vibration images are fully integrated. Through multi-level connections and residual connections, the model's learning and expression capabilities for features are enhanced to obtain the final classification results. The advantages of deep learning and physical models are combined to improve the accuracy and reliability of predictions. At the same time, the comparative learning mechanism is used to reduce the demand for a large amount of labeled data, so that the model can learn effective feature representations under limited data conditions. Multi-level feature fusion is used to enhance the model's adaptability to different working conditions. By optimizing the network structure, the computational complexity of the model is reduced, and the feasibility of practical application is improved, thereby solving the technical problem of low accuracy in life prediction of the input shaft bearing of the gearbox. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0056] Figure 1 This is a flow chart of Embodiment 1 of the method for predicting the life of the input shaft bearing of the gearbox of the present application;

[0057] Figure 2A schematic diagram of SDP image conversion of the gearbox input shaft bearing life prediction method provided in Example 1 of the present application;

[0058] Figure 3 A schematic diagram of feature fusion of a method for predicting the life of a gearbox input shaft bearing provided in Example 1 of the present application;

[0059] Figure 4 A schematic diagram of the CAFFC module structure of the gearbox input shaft bearing life prediction method provided in Example 1 of the present application;

[0060] Figure 5 This is a flow chart of Embodiment 2 of the method for predicting the life of the input shaft bearing of the gearbox of the present application;

[0061] Figure 6 A schematic diagram of a brief process of a method for predicting the life of a gearbox input shaft bearing provided in Embodiment 2 of the present application;

[0062] Figure 7 This is a schematic diagram of the module structure of the gearbox input shaft bearing life prediction device according to an embodiment of the present application;

[0063] Figure 8 Schematic diagram of the equipment structure of the hardware operating environment involved in the method for predicting the life of the gearbox input shaft bearing in the embodiment of the present application.

[0064] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0065] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0066] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0067] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a gearbox input shaft bearing life prediction device, etc., and this embodiment does not specifically limit this. The following takes the gearbox input shaft bearing life prediction device as an example to illustrate this embodiment and the following embodiments.

[0068] The present application embodiment provides a method for predicting the life of a gearbox input shaft bearing, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the method for predicting the life of a gearbox input shaft bearing of the present application.

[0069] In this embodiment, the method for predicting the life of the gearbox input shaft bearing includes steps S10 to S70:

[0070] Step S10, obtaining vibration signals of the gearbox input shaft bearing at different times during the entire life cycle, converting the segmented vibration signals into corresponding vibration images, and extracting sample vibration images from the vibration images;

[0071] It should be noted that this embodiment collects vibration signals at different times during the entire life cycle of the gearbox input shaft bearing for gearbox input shaft bearing life prediction. The vibration signal data can be angular acceleration data collected by a sensor, and can also be flexibly adjusted according to actual conditions, and there is no specific limitation on this.

[0072] In addition, it should be noted that the collected vibration signals need to be processed in segments, and the length of the segments can be set according to actual needs. There is no specific limitation on this. In this embodiment, the collected vibration signals are segmented according to the length of 2560.

[0073] It is understandable that the segmented vibration signal can be converted into a vibration image by using an SDP conversion method, or other suitable conversion methods can be used, which are not specifically limited.

[0074] It should be understood that it is necessary to extract data from the vibration image as sample data for predicting life, that is, sample vibration image. Random sampling can usually be used for extraction, and it can also be flexibly adjusted according to actual needs, without specific restrictions. The number of extractions is usually μN, where N is the total number of vibration images and μ is the set extraction ratio. The specific value of the extraction ratio can be set according to actual conditions, without specific restrictions. Thus, a certain number of vibration signals are extracted as sample vibration images.

[0075] In a feasible implementation manner, the segmented vibration signal is converted into a vibration image using an SDP conversion method.

[0076] It should be noted that the vibration image obtained by conversion at this time is an SDP image.

[0077] It can be understood that, for the two moments with a time interval of l, the corresponding amplitudes of the angular acceleration of the gearbox input shaft bearing are D n , D n+l Through SDP transformation, the angular acceleration amplitude D of the gearbox input shaft bearing corresponding to time n can be nConverted into a point in polar coordinates, recorded as P{r(n), δ(n), φ(n)}, where r(n) is the radius of point P in polar coordinates, that is, the distance between point P and the pole O, called the polar radius of point P; δ(n) is the angle of point P rotated counterclockwise along the polar axis OX, and φ(n) is the angle of point P rotated clockwise along the polar axis OX, called the polar angle of point P, such as Figure 2 shown.

[0078] It can be seen that the characteristic parameters of the polar diameter and polar angle jointly determine the position of point P in polar coordinates, satisfying the calculation relationship:

[0079]

[0080] Among them, D min is the minimum amplitude of the gearbox input shaft bearing angular acceleration in the time series, D max is the maximum amplitude of the angular acceleration of the gearbox input shaft bearing of the time series, θ is the mirror symmetry plane rotation angle, l is the time interval, and κ is the angle magnification factor. It is usually necessary to reasonably select the initial value of the mirror symmetry plane rotation angle θ to avoid excessive image overlap or insufficient symmetry. The value of the angle magnification factor κ can be smaller than the mirror symmetry plane rotation angle θ. The time series l takes values ​​in the range of 1 to 10, and the subtle differences between the time series can be reflected by its different values. Under different states, the parameter setting when the vibration signal is converted to SDP has a certain influence on the properties of the converted image. Specific parameters can be set according to actual conditions to ensure the SDP image quality of the vibration signal. In this embodiment, the mirror symmetry plane rotation angle θ can be set to 60°, the angle magnification factor κ can be set to 40°, and the time interval l can be set to 5. Based on this, the SDP image (vibration image) of the vibration signal of the gearbox input shaft bearing is obtained. The transformation of the frequency components in the vibration signal will be reflected in the SDP image, and different vibration signals can be distinguished.

[0081] The original vibration signal is converted into structured data (SDP image) through the signal processing unit, which not only retains the key information of the signal but also facilitates subsequent data analysis and processing.

[0082] Step S20, determining a physical model life index of the gearbox input shaft bearing based on the life prediction physical model of the gearbox input shaft bearing;

[0083] It should be noted that the life prediction physical model of the gearbox input shaft bearing is a physical model for predicting the life of the gearbox input shaft bearing. The physical model life index focuses on the performance parameters of the gearbox input shaft bearing, and the service life prediction physical model is calculated to evaluate the life of the gearbox input shaft bearing. Generally speaking, the gearbox input shaft bearing is usually a rolling bearing.

[0084] In a feasible implementation, the step of constructing a physical model life index of a gearbox input shaft bearing may include: obtaining performance parameters of the gearbox input shaft bearing; and constructing a life prediction physical model for the gearbox input shaft bearing based on a first corresponding relationship between the performance parameters and the physical model life index.

[0085] It should be noted that the life of the gearbox input shaft bearing usually needs to comprehensively consider the design, operation and performance evaluation requirements of the gearbox input shaft bearing, including multiple influencing factors such as workload, operating conditions, lubrication method, lubrication condition, material properties, geometric structure, etc. This embodiment uses performance parameters to construct a life prediction physical model, and the selected performance parameters at least include rated dynamic load, actual dynamic load, life index, engine speed, reliability correction factor and life correction factor. These parameters can provide comprehensive input data for the life prediction and performance evaluation of the gearbox input shaft bearing.

[0086] It can be understood that the first corresponding relationship between the performance parameter and the physical model life index, that is, the calculation relationship satisfied by the physical model life index, is as follows:

[0087]

[0088] Where L2 is the life index of the physical model, in hours, C d is the rated dynamic load of the gearbox input shaft bearing, F equ is the actual dynamic load of the gearbox input shaft bearing, ε is the life index, n is the engine speed, unit: revolutions per minute (r / min), β1 is the reliability correction factor of the gearbox input shaft bearing, and β2 is the life correction factor of the gearbox input shaft bearing.

[0089] Actual dynamic load F of the gearbox input shaft bearing equ , refers to the load borne by the bearing under actual working conditions, which can represent the stress of the gearbox input shaft bearing and affect the fatigue life of the gearbox input shaft bearing material. Formula F can be used equ =X·F r +Y·F a Calculate, where F r is the radial load, F a is the axial load, X is the radial load factor, and Y is the axial load factor. The rated dynamic load C of the gearbox input shaft bearing dIt is a theoretical value determined by the bearing manufacturer based on the bearing type, bearing structure size, material properties and process design, indicating the maximum load that the bearing can withstand under specific conditions without permanent deformation of the bearing, and is provided in the bearing product catalog or technical documentation. The life index ε is a constant related to the material properties of the bearing and is usually provided by the bearing manufacturer. The value of the life index may be different for different types of rolling bearings. For example, the life index of ball bearings is usually 3, and the life index of roller bearings is usually 10 / 3, depending on the type and size of the roller, and is not specifically limited. The engine speed n usually affects the operating state of the gearbox input shaft bearing, and affects the average pressure and lubrication state of the gearbox input shaft bearing.

[0090] The reliability correction factor β1 of the gearbox input shaft bearing is related to reliability and is usually provided by the bearing manufacturer. The specific value can be referred to Table 1 (for example only, other appropriate values ​​can also be set). Reliability 90% means that a group of sufficiently large and identical bearings are running under the same conditions, and 90% of the bearings do not experience fatigue failure. The fatigue life expected to be reached or exceeded is also called the rated life L. 10 Correspondingly, reliability 95% means that a group of sufficiently large and identical bearings are running under the same conditions, and 95% of the bearings do not fail due to fatigue. The fatigue life is expected to be reached or exceeded, also known as the rated life L5. The rated life L is usually used. 10 As a measure of bearing life, define It can be used to determine the bearing size that is sufficient to avoid fatigue failure. For a given load, the bearing size calculated using high reliability (e.g. 99%) and reliability correction factor will result in the selected bearing being too large. The higher the reliability, the larger the size of the selected bearing. The reliability and reliability correction factor need to be flexibly adjusted according to the actual application. The life correction factor β2 of the gearbox input shaft bearing is obtained based on the relationship curve between the lubricant dynamic viscosity ratio, lubricant contamination coefficient, and fatigue load limit value of the gearbox input shaft bearing, and is provided by the bearing manufacturer.

[0091] Table 1

[0092] reliability(%) Rated life <![CDATA[Reliability correction factor β1]]> 90 <![CDATA[L 10 ]]> 1 95 <![CDATA[L5]]> 0.62 96 <![CDATA[L4]]> 0.53 97 <![CDATA[L3]]> 0.44 98 <![CDATA[L2]]> 0.33 99 <![CDATA[L1]]> 0.21

[0093] It should be understood that the actual bearing life is affected by a combination of factors, such as bearing type, lubrication conditions, bearing installation accuracy, fatigue characteristics of bearing materials, dynamic characteristics of crankshafts, ignition order of multi-cylinder engines, bearing geometry, lubrication conditions, bearing clearance, material characteristics, environmental factors, etc. Therefore, in practical applications, when constructing a physical model for predicting the life of the gearbox input shaft bearing, factors such as bearing fatigue life, wear life, and lubrication life can also be considered, and the calculation relationship used in the model can be flexibly adjusted according to actual conditions to ensure the reliability and performance optimization of the gearbox input shaft bearing design.

[0094] Step S30, annotating a preset number of sample vibration images based on the physical model life index to obtain annotated vibration images and unannotated vibration images;

[0095] It should be noted that, in this embodiment, a small number of sample vibration images are annotated using the physical model life index, and the annotated vibration images are annotated sample vibration images, and the unannotated vibration images are unannotated sample vibration images.

[0096] In addition, it should be noted that the preset number, i.e. the number of labeled samples, is usually small and can be flexibly adjusted according to actual needs. There is no specific limitation on this. The number of labeled vibration images is much smaller than the number of unlabeled vibration images.

[0097] Step S40, inputting the annotated vibration image and the unannotated vibration image into a dual-branch network for feature extraction, and obtaining corresponding feature representation vectors respectively;

[0098] It can be understood that the present embodiment uses a dual-branch network to perform multi-feature extraction on the labeled vibration image and the unlabeled vibration image, and the two types of vibration images are respectively input into their respective networks for training and extraction of corresponding features.

[0099] It should be noted that the dual-branch network includes a first branch network and a second branch network, and the first branch network and the second branch network both include a coding layer, a pooling layer, and a projection head connected in sequence. The feature representation vector includes a first feature representation vector and a second feature representation vector, the first feature representation vector is a vector composed of features obtained by extracting features from an unlabeled vibration image using the first branch network, and the second feature representation vector is a vector composed of features obtained by extracting features from a labeled vibration image using the second branch network.

[0100] In a feasible implementation, step S40 may include: inputting the dimensionally reduced unlabeled vibration image into the encoding layer of the first branch network for feature extraction to obtain a first initial feature, and inputting the first initial feature into the pooling layer and projection head of the first branch network to obtain a first feature representation vector of the unlabeled vibration image; inputting the labeled vibration image into the encoding layer of the second branch network for feature extraction to obtain a second initial feature, and inputting the second initial feature into the pooling layer and projection head of the second branch network to obtain a second feature representation vector of the labeled vibration image.

[0101] It should be noted that the annotated vibration image can be recorded as h×w represents the length and width of the annotated vibration image, c p represents the number of channels of the data. The unlabeled vibration image can be denoted as c o The number of channels representing the data. Considering the redundant information in the unlabeled vibration image, in order to reduce the computational cost, it is necessary to perform dimensionality reduction processing on the unlabeled vibration image. The method of dimensionality reduction can be flexibly adjusted according to the actual situation, and there is no specific limitation on this. For example, principal component analysis is used to perform dimensionality reduction processing on the unlabeled vibration image, that is, D o Perform dimensionality reduction to obtain c oo It is the number of channels after dimensionality reduction, which is set according to the actual situation and is not specifically limited.

[0102] It can be understood that the unlabeled vibration image after dimensionality reduction uses the encoding layer of the first branch network to extract features, and the obtained features are further input into the pooling layer and the projection head for processing to obtain the first feature representation vector z1. The labeled vibration image uses the encoding layer of the second branch network to extract features, and the obtained features are further input into the pooling layer and the projection head for processing to obtain the second feature representation vector z2.

[0103] In this embodiment, the coding layer adopts the structure of convolution layer + normalization layer + activation layer, and the number of convolution layers, convolution type, number and size of convolution kernels and other parameters are set according to the actual situation without specific limitation. For example, two-dimensional convolution can be used, and four convolution layers are used. The parameters of these four convolution layers are 32@3*3, 64@3*3, 128@3*3, and 256@3*3, that is, the number of convolution kernels is 32, 64, 128, and 256, and the size is 3*3.

[0104] Step S50, calculating the evaluation index of the feature representation vector, and determining the comprehensive evaluation index based on the evaluation index, wherein the evaluation index at least includes a feature consistency index, a separability index, a robustness index, an overall feature consistency index, and a fusion quality index;

[0105] It should be noted that the initially obtained feature representation vector needs to calculate the corresponding evaluation indicators, which include at least feature consistency indicator, separability indicator, robustness indicator, overall feature consistency indicator and fusion quality indicator.

[0106] The feature consistency index is an index for evaluating feature consistency. The feature consistency index of the first feature characterization vector and the second feature characterization vector can be calculated using the following formula:

[0107]

[0108] In the formula, σ1 represents the feature consistency index of the first feature representation vector, z 1,k represents the kth feature in the first feature representation vector, represents the feature mean of the first feature representation vector, d represents the feature dimension, σ2 represents the feature consistency index of the second feature representation vector, z 2,k represents the kth feature in the second feature representation vector, Represents the feature mean of the second feature characterization vector. Substituting the relevant data, the feature consistency index of the first feature characterization vector and the second feature characterization vector can be calculated.

[0109] The separability index is an index for evaluating the separability of features. This embodiment uses an inter-class method to measure the degree of separation of features between different classes and calculate the separability index. The separability index of the first feature characterization vector and the second feature characterization vector can be calculated using the following calculation formula:

[0110]

[0111] In the formula, represents the separability index of the first eigenvector, represents the separability index of the second feature representation vector, C represents the number of categories, n c represents the number of samples of category c, represents the feature mean of the first feature representation vector category c, represents the feature mean of the second feature representation vector category c, represents the eigenvalue of the first eigenvector, Represents the feature mean of the second feature characterization vector. Substituting the relevant data, the separability index of the first feature characterization vector and the second feature characterization vector can be calculated.

[0112] The robustness index is an index for evaluating robustness. In this embodiment, feature robustness is evaluated by adding noise. Feature robustness can be calculated using the following relationship:

[0113]

[0114] Where R1 represents the robustness index of the first feature representation vector, R2 represents the robustness index of the second feature representation vector, and z 1,k represents the kth feature in the first feature representation vector, z 2,k represents the kth feature in the second feature representation vector, d represents the feature dimension, z1′ ,k represents the feature obtained after noise perturbation of the kth feature in the first feature representation vector, z2′ ,k It represents the feature obtained after the kth feature in the second feature representation vector is disturbed by noise. Substituting the relevant data, the robustness index of the first feature representation vector and the second feature representation vector can be calculated.

[0115] In a feasible implementation, the step of calculating the overall feature consistency index includes: obtaining the third correspondence between the feature characterization vector, the feature mean and the overall feature consistency index; determining the overall feature consistency index based on the first feature characterization vector, the feature mean of the first feature characterization vector, the second feature characterization vector, the feature mean of the second feature characterization vector and the third correspondence.

[0116] It should be noted that the overall feature consistency index is an index for evaluating the consistency between features. The third corresponding relationship between the feature representation vector, the feature mean and the overall feature consistency index is the calculation relationship of the overall feature consistency index, as shown below:

[0117]

[0118] In the formula, r represents the overall feature consistency index, z 1,k represents the kth feature in the first feature representation vector, z 2,k represents the kth feature in the second feature representation vector, d represents the number of feature dimensions, represents the eigenvalue of the first eigenvector, Represents the feature mean of the second feature representation vector. Substituting the relevant data, the overall feature consistency index can be calculated.

[0119] In a feasible implementation, the step of calculating the fusion quality index includes: obtaining a fourth correspondence between a feature characterization vector, a feature dimension number and a fusion quality index; and determining the fusion quality index based on the first feature characterization vector, the second feature characterization vector, the feature dimension number and the fourth correspondence.

[0120] It can be understood that the fusion quality index is an index for evaluating the fusion quality of features. The fourth corresponding relationship between the feature representation vector, the feature dimension number and the fusion quality index is the calculation relationship of the fusion quality index, as shown below:

[0121]

[0122] In the formula, z 1,k represents the kth feature in the first feature representation vector, z 2,k represents the kth feature in the second feature representation vector, and d represents the feature dimension. Substituting the relevant data, the fusion quality index can be calculated.

[0123] In a feasible implementation, based on the evaluation index, the step of determining the comprehensive evaluation index may include steps S501 to S504:

[0124] Step S501, calculating the index score of the first branch network based on the feature consistency index, the separability index, the robustness index, the weight of the feature consistency index, the weight of the separability index and the weight of the robustness index corresponding to the first feature representation vector;

[0125] It should be noted that the index score of the first branch network can be calculated using the following calculation formula:

[0126]

[0127] In the formula, S1 represents the index score of the first branch network, σ1 represents the feature consistency index of the first feature representation vector, represents the separability index of the first feature representation vector, R1 represents the robustness index of the first feature representation vector, γ1 represents the weight of the feature consistency index, γ2 represents the weight of the separability index, and γ3 represents the weight of the robustness index, where the specific value of the weight can be set according to actual needs and is not specifically limited. Substituting the relevant data, the index score of the first branch network can be calculated.

[0128] Step S502, calculating the index score of the second branch network based on the feature consistency index, the separability index, the robustness index, the weight of the feature consistency index, the weight of the separability index and the weight of the robustness index corresponding to the second feature representation vector;

[0129] It should be noted that the index score of the second branch network can be calculated using the following calculation formula:

[0130]

[0131] In the formula, S2 represents the index score of the second branch network, σ2 represents the feature consistency index of the second feature representation vector, represents the separability index of the second feature representation vector, R2 represents the robustness index of the second feature representation vector, γ1 represents the weight of the feature consistency index, γ2 represents the weight of the separability index, and γ3 represents the weight of the robustness index, where the specific value of the weight can be set according to actual needs and is not specifically limited. Substituting the relevant data, the index score of the second branch network can be calculated.

[0132] Step S503, obtaining a second corresponding relationship between the overall feature consistency index, the fusion quality index, the weight of the overall feature consistency index, the weight of the fusion quality index, the index score, the balance coefficient and the comprehensive evaluation index;

[0133] It should be noted that the second corresponding relationship between the overall feature consistency index, the fusion quality index, the weight of the overall feature consistency index, the weight of the fusion quality index, the index score, the balance coefficient and the comprehensive evaluation index, that is, the calculation relationship of the comprehensive evaluation index, is as follows:

[0134] H=β1r+β2Q+αS1+(1-α)S2

[0135] In the formula, H represents the comprehensive evaluation index, S1 represents the index score of the first branch network, S2 represents the index score of the second branch network, r represents the overall feature consistency index, β1 represents the weight of the overall feature consistency index, β2 represents the weight of the fusion quality index, Q represents the fusion quality index, α represents the balance coefficient, which is a coefficient used to balance the index scores of the two branch networks. The specific values ​​of the weights and the balance coefficients can be set according to actual needs and are not specifically limited.

[0136] Step S504, determining a comprehensive evaluation index based on the overall feature consistency index, the fusion quality index, the weight of the overall feature consistency index, the weight of the fusion quality index, the index score of the first branch network, the index score of the second branch network, the balance coefficient and the second corresponding relationship.

[0137] It can be understood that by substituting the relevant data into the above second corresponding relationship, the corresponding comprehensive evaluation index can be calculated.

[0138] Step S60, when the comprehensive evaluation index is greater than or equal to the evaluation index threshold, the feature representation vector is fused to obtain a fused feature;

[0139] It should be noted that the evaluation index threshold is the threshold of the set comprehensive evaluation index. The specific value can be flexibly adjusted according to actual needs and is not specifically limited.

[0140] It is understandable that after obtaining the comprehensive evaluation index, it is necessary to decide whether to perform feature fusion based on the comprehensive evaluation index. When the comprehensive evaluation index is greater than or equal to the evaluation index threshold, it is considered that the next feature fusion can be performed, and then the feature fusion stage is entered to fuse the first feature representation vector and the second feature representation vector to obtain the fused feature, that is, the fused feature. When the comprehensive evaluation index is less than the evaluation index threshold, it is considered that the next feature fusion cannot be performed, and the feature extraction training process is repeated, the parameters of the dual-branch network are updated, and feature extraction is repeated.

[0141] In a feasible implementation, when the comprehensive evaluation index is less than the evaluation index threshold, the first feature characterization vector is input into the prediction layer of the first branch network to obtain the first prediction data; based on the first prediction data, the second prediction data corresponding to the second feature characterization vector is symmetrically determined; the first correlation coefficient between the second prediction data and the first feature characterization vector is calculated, and based on the gradient stopping strategy of the second branch network, the second correlation coefficient between the first prediction data and the second feature characterization vector is calculated; based on the first correlation coefficient, the second correlation coefficient and the correlation factor, the contrast loss function of the dual-branch network is determined; based on the contrast loss function of the dual-branch network and the gradient stopping strategy of the second branch network, the parameters of the dual-branch network are updated, and the labeled vibration image and the unlabeled vibration image are input into the dual-branch network for feature extraction to obtain the corresponding feature characterization vectors respectively.

[0142] It should be noted that the first branch network also includes a prediction layer connected to the projection head. The prediction layer of the first branch network is used to predict the first feature representation vector to obtain the corresponding prediction data, that is, the first prediction data p1, and then the prediction is performed symmetrically to obtain the second prediction data p2 corresponding to the second feature representation vector. In order to learn more about the spatial distribution, data similarity and discriminant feature representation between the two state data in contrastive learning, and to predict the life more accurately, this embodiment defines the contrast loss function as follows:

[0143] L=-(ω1r(p2,z1)+ω2r(p1,stop-gradient(z2)))

[0144] Where L represents the contrast loss function, ω1 and ω2 represent the correlation factors of the contrast loss function, r(*,*) represents the correlation coefficient, stop-gradient represents the gradient stop strategy, z1 represents the first feature representation vector, z2 represents the second feature representation vector, p1 represents the first predicted data, and p2 represents the second predicted data. The specific values ​​of the correlation factors of the contrast loss function can be set according to the actual situation, for example, ω1 and ω2 are set to 0.5 and 0.5 respectively.

[0145] It can be seen that the first correlation coefficient between the second predicted data and the first feature representation vector, as well as the second correlation coefficient between the first predicted data and the second feature representation vector under the gradient stop strategy can be calculated first, and then the first correlation coefficient, the second correlation coefficient and the set correlation factor are substituted into the above formula to calculate the current contrast loss function. Among them, the calculation relationship of the first correlation coefficient is as follows:

[0146]

[0147] Wherein, r(p2,z1) represents the first correlation coefficient between the second prediction data p2 and the first feature representation vector z1, z 1,k represents the kth feature in the first feature representation vector, represents the feature mean of the first feature representation vector, p 2,k represents the predicted data of the kth feature, represents the mean of the second prediction data, and d represents the number of feature dimensions.

[0148] This embodiment applies a gradient stop strategy to z2, that is, the gradient is not calculated for the conversion process from z2 to p2 during back propagation, ensuring that the feature extractor parameters of z2 are not updated, and only the parameters of the prediction head p2 are updated. Based on the contrast loss function, the second branch network adopts a gradient stop strategy to update the parameters of the dual-branch network, thereby returning to step S40, re-extracting features from the labeled vibration image and the unlabeled vibration image, and re-deciding whether to perform feature fusion.

[0149] It should be understood that the use of gradient stopping strategy helps the model learn stable and useful feature representations in the pre-training stage, while laying a good foundation for subsequent downstream stages such as feature fusion. In addition, through the contrastive learning mechanism, the dependence on a large amount of labeled data can be reduced, so that the model can learn effective feature representations under limited data conditions, which is particularly valuable for actual industrial applications where data labeling costs are high.

[0150] In a feasible implementation manner, the step of fusing the feature representation vectors to obtain the fused features includes steps S601 to S603:

[0151] Step S601, performing a residual operation on the feature representation vector to obtain a residual feature;

[0152] It should be noted that the reference Figure 3 , assuming that the encoding layer contains four convolutional layers, the unlabeled vibration image D p The features obtained after four convolutional layers are: Where i = 1, 2, 3, 4, It is the convolutional layer of the upper network (the first branch network), and the annotated vibration image D after dimensionality reductiono ′The features obtained after four convolutional layers are: Where i = 1, 2, 3, 4, It is the convolution layer of the lower network (the second branch network). Then the residual operation is performed to obtain the residual features of the two branch networks, namely p′4=p1+p4 and o4′=o1+o4.

[0153] Step S602, performing multi-level connection on the weighted residual features to obtain initial fusion features;

[0154] It can be understood that the residual features of the two branch networks are weighted according to the weights, and the initial fusion features are obtained through multi-level connections, that is, y j =δ p *p j ' +1 +δ o *o j ' +1 ,j=1,2,3, where δ p , δ o are two learnable parameter tensors that can be automatically updated by the optimizer, allowing the model to adaptively add the features of the two modalities.

[0155] Step S603: input the initial fusion features into the feature fusion classification module to obtain fusion features.

[0156] It should be noted that the reference Figure 3 , get the initial fusion feature y h It needs to pass through the feature fusion classification module CAFFC with the same structure in sequence to obtain the fused features.

[0157] Understandably, reference Figure 4 The feature fusion classification module includes the first convolution layer, the normalization layer, the first activation layer, the channel attention layer, the pooling layer, the second convolution layer and the second activation layer connected in sequence. The initial fusion feature y j The first convolutional layer is input and operated with a convolution kernel size of 1×1, which can transform features in the channel dimension. The output of the first convolutional layer is normalized by a normalization layer to accelerate convergence and improve model stability. The nonlinear activation function ReLU is introduced in the first activation layer to increase the nonlinear expression ability of the model. In the channel attention layer, the global spatial information of each channel is obtained through global average pooling, and then a fully connected layer and Softmax function are used to assign weights to each channel to achieve adaptive weighting of feature channels and obtain feature y att , the calculation relationship is as follows:

[0158] F=Conv 1*1 (y)

[0159] G = softmax(F)

[0160] y att =y⊙G

[0161] Where ⊙ represents the channel-by-channel product, y att is the adaptively weighted feature. Feature y att Entering the pooling layer, this embodiment uses an average pooling operation to reduce the feature size to 1×1 to prepare for the subsequent life estimation operation. Then in the second convolution layer, the 1×1 feature output by the pooling layer is further reduced from the original dimension to the dimension of the number of categories C, and finally the initial fusion features y are obtained through the second activation layer. j Output Through the forward propagation process, each initial fusion feature y j Three outputs can be obtained

[0162] In getting the output After that, the loss function is used to calculate each input y j The loss value L j ,Right now The loss function Loss() can be selected according to the actual situation. For example, this embodiment adopts the cross entropy loss function. L1, L2, and L3 supervise the learning process of low-level, middle-level, and high-level fusion features respectively. The final overall loss value L can be expressed as:

[0163] L=θ1L1+θ2L2+θ3L3

[0164] Among them, θ i is the adaptive weight, and the calculation relationship is:

[0165]

[0166] Therefore, the overall loss value L can be optimized using back propagation, and then the parameters of the feature fusion classification module CAFFC are updated, and then the trained CAFFC can be used to obtain the final fusion feature.

[0167] It should be understood that the adoption of a multi-level feature fusion strategy fully integrates the multimodal heterogeneous features of the bearing vibration image, which can enhance the model's adaptability and recognition ability to bearing state changes under different working conditions. Using multi-level connections and residual connections, the model can learn and express complex features more deeply, thereby improving the accuracy and robustness of the classification results. The optimized network structure design reduces the computational complexity of the model, which not only speeds up the training and reasoning speed of the model, but also reduces the demand for computing resources, making the model more suitable for deployment and application in resource-constrained industrial environments, improving the feasibility of the model in actual industrial scenarios, and making the predictive maintenance of bearings more efficient and economical.

[0168] Step S70, determining a life prediction result of the gearbox input shaft bearing based on the fusion features.

[0169] It should be noted that the final life prediction result is obtained by using the fusion features. The life prediction result is usually the predicted remaining life, which can be expressed in the form of a numerical value, for example: 75%, 80%.

[0170] The present embodiment provides a method for predicting the life of a gearbox input shaft bearing, which converts a vibration signal into a vibration image, then extracts part of the vibration images as sample data for preliminary life prediction, compares and studies a small number of vibration images labeled with physical model life indicators and a large number of unlabeled vibration images, and fully integrates the multimodal heterogeneous features of the bearing vibration images. Through multi-level connections and residual connections, the model's learning and expression capabilities for features are enhanced to obtain the final classification results. The advantages of deep learning and physical models are combined to improve the accuracy and reliability of prediction. At the same time, the comparative learning mechanism is used to reduce the demand for a large amount of labeled data, so that the model can learn effective feature representation under limited data conditions. Multi-level feature fusion is used to enhance the model's adaptability to different working conditions. By optimizing the network structure, the computational complexity of the model is reduced, and the feasibility of practical application is improved.

[0171] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 5 , step S70 may include step S80:

[0172] Step S80, determining maintenance recommendations based on the life prediction results of the gearbox input shaft bearing.

[0173] It should be noted that after obtaining the life prediction results of the gearbox input shaft bearing, relative maintenance suggestions are given based on the prediction results.

[0174] When the remaining life is greater than or equal to 75%, maintenance recommendations usually include preventive maintenance and performance monitoring. At this stage, the life of the bearing is still long, and the focus should be on preventive maintenance, such as regular lubrication, checking the cleanliness and lubrication condition of the bearing, and ensuring there are no signs of wear or damage. In addition, start implementing or strengthening bearing performance monitoring, recording key indicators such as bearing temperature and vibration, so as to detect potential problems early.

[0175] When the remaining life is greater than or equal to 50% and less than 75%, maintenance recommendations usually include two aspects: enhanced performance monitoring and bearing replacement planning. For enhanced performance monitoring, you can increase the monitoring frequency and use more advanced monitoring technologies, such as online vibration analysis, to more accurately predict the wear of bearings. For bearing replacement planning, you can start to consider developing a bearing replacement plan, including determining the replacement time, budget, and required resources.

[0176] When the remaining life is greater than or equal to 25% and less than 50%, maintenance recommendations usually include regular inspection and design improvement. For regular inspection, the inspection cycle can be shortened and the condition of the bearing can be checked more frequently, including checking the bearing wear, lubrication and working environment. For design improvement, design improvements to the bearing or its working environment can be considered to extend its life.

[0177] When the remaining life is less than 25%, maintenance recommendations usually include emergency replacement plans and spare parts inventory. An emergency replacement plan can be developed to ensure that the bearing can be replaced in time before the life is exhausted. In addition, ensure that there is enough spare parts inventory to quickly replace the bearing when it fails to reduce downtime.

[0178] This embodiment provides a method for predicting the life of a gearbox input shaft bearing, which converts a vibration signal into a vibration image, then extracts part of the vibration images as sample data for preliminary life prediction, compares and studies a small number of vibration images labeled with physical model life indicators and a large number of unlabeled vibration images, and fully integrates the multimodal heterogeneous features of the bearing vibration images. Through multi-level connections and residual connections, the model's learning and expression capabilities for features are enhanced to obtain the final classification results, and relative maintenance recommendations are given based on the results to ensure the safe use of the gearbox input shaft bearings.

[0179] For example, in order to help understand the implementation process of the gearbox input shaft bearing life prediction method obtained by combining this embodiment with the above-mentioned embodiment 2, please refer to Figure 6 , Figure 6 A brief flow chart of a gearbox input shaft bearing life prediction method is provided, specifically:

[0180] Step 1: Data preparation: collect vibration signals of the gearbox input shaft bearing at different times during its life cycle and divide them into segments;

[0181] Step 2: The signal processing unit converts the collected vibration signal into a vibration image;

[0182] Step 3: The data extraction unit extracts samples from a large number of vibration images by random sampling;

[0183] Step 4: Input the extracted vibration image into the physical life analysis unit for preliminary life prediction;

[0184] Step 5: Use a small number of vibration images annotated with the physical model life index and a large number of unlabeled vibration images for comparative learning, input them into the dual-branch network, and use the encoding layer of each network to extract multiple features of the vibration signal;

[0185] Step 6: After passing the features of the two types of data through the pooling layer and the projection head, two types of feature representation vectors are obtained;

[0186] Step 7: Input the two types of feature representation vectors initially obtained into the feature fusion decision unit for fusion judgment to see whether feature fusion can be performed. If the comprehensive evaluation index is greater than or equal to the evaluation index threshold, the feature fusion stage is entered. If it is not satisfied, the feature extraction training process is repeated and feature extraction is repeated.

[0187] Step 8: Perform feature fusion process and determine the final life prediction result based on the obtained fusion features.

[0188] This application also provides a gearbox input shaft bearing life prediction device, please refer to Figure 7 , the gearbox input shaft bearing life prediction device includes:

[0189] The feature extraction module 10 is used to obtain the vibration signals of the gearbox input shaft bearing at different times during the whole life cycle, convert the segmented vibration signals into corresponding vibration images, and extract sample vibration images from the vibration images;

[0190] The feature extraction module 10 is also used to determine the physical model life index of the gearbox input shaft bearing based on the life prediction physical model of the gearbox input shaft bearing;

[0191] The feature extraction module 10 is also used to annotate a preset number of sample vibration images based on the physical model life index to obtain annotated vibration images and unannotated vibration images;

[0192] The feature extraction module 10 is also used to input the labeled vibration image and the unlabeled vibration image into the dual-branch network for feature extraction, and obtain corresponding feature representation vectors respectively;

[0193] A feature fusion module 20 is used to calculate the evaluation index of the feature representation vector, and determine the comprehensive evaluation index based on the evaluation index, where the evaluation index at least includes a feature consistency index, a separability index, a robustness index, an overall feature consistency index, and a fusion quality index;

[0194] The feature fusion module 20 is further used to fuse the feature representation vectors to obtain fused features when the comprehensive evaluation index is greater than or equal to the evaluation index threshold;

[0195] The life prediction module 30 is used to determine the life prediction result of the gearbox input shaft bearing based on the fusion feature.

[0196] In a feasible implementation manner, the feature extraction module 10 is further used to obtain performance parameters of the gearbox input shaft bearing, the performance parameters at least including rated dynamic load, actual dynamic load, life index, engine speed, reliability correction factor and life correction factor;

[0197] Based on the first correspondence between the performance parameters and the life indicators of the physical model, a life prediction physical model for the gearbox input shaft bearing is constructed.

[0198] In a feasible implementation manner, the feature characterization vector includes a first feature characterization vector and a second feature characterization vector, the dual-branch network includes a first branch network and a second branch network, the first branch network and the second branch network both include a coding layer, a pooling layer, and a projection head connected in sequence, and the feature extraction module 10 is further used to input the dimensionally reduced unlabeled vibration image into the coding layer of the first branch network to perform feature extraction to obtain a first initial feature, and input the first initial feature into the pooling layer and the projection head of the first branch network to obtain a first feature characterization vector of the unlabeled vibration image;

[0199] The annotated vibration image is input into the encoding layer of the second branch network for feature extraction to obtain a second initial feature, and the second initial feature is input into the pooling layer and the projection head of the second branch network to obtain a second feature representation vector of the annotated vibration image.

[0200] In a feasible implementation manner, the feature fusion module 20 is further used to calculate the index score of the first branch network based on the feature consistency index, the separability index, the robustness index, the weight of the feature consistency index, the weight of the separability index, and the weight of the robustness index corresponding to the first feature representation vector;

[0201] Calculate the index score of the second branch network based on the feature consistency index, the separability index, the robustness index, the weight of the feature consistency index, the weight of the separability index, and the weight of the robustness index corresponding to the second feature representation vector;

[0202] Obtaining a second corresponding relationship between the overall feature consistency index, the fusion quality index, the weight of the overall feature consistency index, the weight of the fusion quality index, the index score, the balance coefficient and the comprehensive evaluation index;

[0203] Based on the overall feature consistency index, the fusion quality index, the weight of the overall feature consistency index, the weight of the fusion quality index, the index score of the first branch network, the index score of the second branch network, the balance coefficient and the second corresponding relationship, a comprehensive evaluation index is determined.

[0204] In a feasible implementation manner, the feature fusion module 20 is further used to obtain a third corresponding relationship between the feature representation vector, the feature mean and the overall feature consistency index;

[0205] Determining an overall feature consistency index based on the first feature characterization vector, the feature mean of the first feature characterization vector, the second feature characterization vector, the feature mean of the second feature characterization vector, and the third corresponding relationship;

[0206] Obtaining a fourth corresponding relationship between the feature representation vector, the feature dimension number and the fusion quality index;

[0207] A fusion quality indicator is determined based on the first feature characterization vector, the second feature characterization vector, the number of feature dimensions, and the fourth corresponding relationship.

[0208] In a feasible implementation manner, the feature fusion module 20 is further configured to input the first feature representation vector into the prediction layer of the first branch network to obtain first prediction data when the comprehensive evaluation index is less than the evaluation index threshold;

[0209] Based on the first predicted data, symmetrically determine second predicted data corresponding to the second feature characterization vector;

[0210] Calculating a first correlation coefficient between the second prediction data and the first feature characterization vector, and calculating a second correlation coefficient between the first prediction data and the second feature characterization vector based on a gradient stopping strategy of the second branch network;

[0211] Determine a contrast loss function of the dual-branch network based on the first correlation coefficient, the second correlation coefficient, and the correlation factor;

[0212] Based on the contrast loss function of the dual-branch network and the gradient stopping strategy of the second branch network, the parameters of the dual-branch network are updated, and the labeled vibration image and the unlabeled vibration image are input into the dual-branch network for feature extraction to obtain the corresponding feature representation vectors.

[0213] In a feasible implementation manner, the feature fusion module 20 is further used to perform residual operation on the feature representation vector to obtain residual features;

[0214] The weighted residual features are connected in multiple levels to obtain the initial fusion features;

[0215] The initial fusion features are input into the feature fusion classification module to obtain the fusion features. The feature fusion classification module includes a first convolution layer, a normalization layer, a first activation layer, a channel attention layer, a pooling layer, a second convolution layer and a second activation layer connected in sequence.

[0216] In a feasible implementation manner, the life prediction module 30 is further used to determine maintenance recommendations based on the life prediction results of the gearbox input shaft bearing.

[0217] The gearbox input shaft bearing life prediction device provided by the present application adopts the gearbox input shaft bearing life prediction method in the above embodiment, which can solve the technical problem of low gearbox input shaft bearing life prediction accuracy. Compared with the prior art, the beneficial effects of the gearbox input shaft bearing life prediction device provided by the present application are the same as the beneficial effects of the gearbox input shaft bearing life prediction method provided by the above embodiment, and other technical features in the gearbox input shaft bearing life prediction device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0218] The present application provides a gearbox input shaft bearing life prediction device, the gearbox input shaft bearing life prediction device includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the gearbox input shaft bearing life prediction method in the above-mentioned embodiment one.

[0219] Reference below Figure 8 , which shows a schematic diagram of a structure of a gearbox input shaft bearing life prediction device suitable for implementing the embodiment of the present application. The gearbox input shaft bearing life prediction device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptops, digital broadcast receivers, personal digital assistants, tablet computers, portable multimedia players, vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 8 The gearbox input shaft bearing life prediction device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0220] like Figure 8As shown, the gearbox input shaft bearing life prediction device may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the gearbox input shaft bearing life prediction device are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems may be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 may allow the gearbox input shaft bearing life prediction device to communicate wirelessly or wired with other devices to exchange data. Although the gearbox input shaft bearing life prediction device having various systems is shown in the figure, it should be understood that it is not required to implement or have all the systems shown.

[0221] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0222] The gearbox input shaft bearing life prediction device provided by the present application adopts the gearbox input shaft bearing life prediction method in the above embodiment, which can solve the technical problem of low gearbox input shaft bearing life prediction accuracy. Compared with the prior art, the gearbox input shaft bearing life prediction device provided by the present application has the same beneficial effects as the gearbox input shaft bearing life prediction method provided by the above embodiment, and other technical features in the gearbox input shaft bearing life prediction device are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0223] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the method for predicting the life of a gearbox input shaft bearing in the above-mentioned embodiment.

[0224] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above. The computer-readable storage medium may be included in the gearbox input shaft bearing life prediction device; or may exist independently without being assembled into the gearbox input shaft bearing life prediction device.

[0225] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the gearbox input shaft bearing life prediction device, the gearbox input shaft bearing life prediction device: obtains vibration signals at different times in the entire life cycle of the gearbox input shaft bearing, converts the segmented vibration signals into corresponding vibration images, and extracts sample vibration images from the vibration images; determines the physical model life index of the gearbox input shaft bearing based on the life prediction physical model of the gearbox input shaft bearing; annotates a preset number of sample vibration images based on the physical model life index to obtain annotated vibration images and unlabeled vibration images; inputs the annotated vibration images and the unlabeled vibration images into a dual-branch network for feature extraction to obtain corresponding feature representation vectors respectively; calculates the evaluation index of the feature representation vector, and determines the comprehensive evaluation index based on the evaluation index; when the comprehensive evaluation index is greater than or equal to the evaluation index threshold, fuses the feature representation vector to obtain a fused feature; and determines the life prediction result of the gearbox input shaft bearing based on the fused feature.

[0226] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code includes one or more executable instructions for realizing the logical function of the specification. It should also be noted that in some implementations as replacements, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings.

[0227] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0228] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned gearbox input shaft bearing life prediction method, and can solve the technical problem of low accuracy in gearbox input shaft bearing life prediction. The beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the gearbox input shaft bearing life prediction method provided in the above-mentioned embodiment, and will not be repeated here.

[0229] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned method for predicting the life of a gearbox input shaft bearing when the computer program is executed by a processor.

[0230] The computer program product provided in this application can solve the technical problem of low accuracy in predicting the life of the input shaft bearing of the gearbox. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the method for predicting the life of the input shaft bearing of the gearbox provided in the above embodiment, which will not be elaborated here.

[0231] The above are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect application in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for predicting the life of a gearbox input shaft bearing, characterized in that: The method comprises: Acquire vibration signals of the gearbox input shaft bearing at different times during the entire life cycle, convert the segmented vibration signals into corresponding vibration images, and extract sample vibration images from the vibration images; Determining a physical model life index of the gearbox input shaft bearing based on the gearbox input shaft bearing life prediction physical model; Based on the physical model life index, a preset number of sample vibration images are labeled to obtain labeled vibration images and unlabeled vibration images; Inputting the labeled vibration image and the unlabeled vibration image into a dual-branch network for feature extraction to obtain corresponding feature representation vectors respectively; Calculating the evaluation index of the feature representation vector, and determining the comprehensive evaluation index based on the evaluation index, wherein the evaluation index at least includes a feature consistency index, a separability index, a robustness index, an overall feature consistency index and a fusion quality index; When the comprehensive evaluation index is greater than or equal to the evaluation index threshold, fusing the feature representation vectors to obtain a fused feature; Based on the fused features, a life prediction result of the gearbox input shaft bearing is determined.

2. The method according to claim 1, characterized in that Before the step of determining the physical model life index of the gearbox input shaft bearing based on the life prediction physical model of the gearbox input shaft bearing, the step further includes: Acquire performance parameters of a gearbox input shaft bearing, wherein the performance parameters at least include a rated dynamic load, an actual dynamic load, a life index, an engine speed, a reliability correction factor, and a life correction factor; Based on the first corresponding relationship between the performance parameter and the physical model life index, a life prediction physical model for the gearbox input shaft bearing is constructed.

3. The method according to claim 1, characterized in that The feature representation vector includes a first feature representation vector and a second feature representation vector, the dual-branch network includes a first branch network and a second branch network, the first branch network and the second branch network each include a coding layer, a pooling layer and a projection head connected in sequence, and the step of inputting the labeled vibration image and the unlabeled vibration image into the dual-branch network for feature extraction to obtain corresponding feature representation vectors respectively includes: Input the dimension-reduced unlabeled vibration image into the encoding layer of the first branch network to perform feature extraction to obtain a first initial feature, and input the first initial feature into the pooling layer and the projection head of the first branch network to obtain a first feature representation vector of the unlabeled vibration image; The annotated vibration image is input into the encoding layer of the second branch network to perform feature extraction to obtain a second initial feature, and the second initial feature is input into the pooling layer and projection head of the second branch network to obtain a second feature representation vector of the annotated vibration image.

4. The method according to claim 3, characterized in that The step of determining the comprehensive evaluation index based on the evaluation index comprises: Calculating an index score of the first branch network based on a feature consistency index, a separability index, a robustness index, a weight of a feature consistency index, a weight of a separability index, and a weight of a robustness index corresponding to the first feature representation vector; Calculating an index score of the second branch network based on the feature consistency index, the separability index, the robustness index, the weight of the feature consistency index, the weight of the separability index, and the weight of the robustness index corresponding to the second feature representation vector; Obtaining a second corresponding relationship between the overall feature consistency index, the fusion quality index, the weight of the overall feature consistency index, the weight of the fusion quality index, the index score, the balance coefficient and the comprehensive evaluation index; The comprehensive evaluation index is determined based on the overall feature consistency index, the fusion quality index, the weight of the overall feature consistency index, the weight of the fusion quality index, the index score of the first branch network, the index score of the second branch network, the balance coefficient and the second corresponding relationship.

5. The method according to claim 4, characterized in that The method further comprises: Obtaining a third correspondence between the feature representation vector, the feature mean, and the overall feature consistency index; Determining the overall feature consistency index based on the first feature characterization vector, the feature mean of the first feature characterization vector, the second feature characterization vector, the feature mean of the second feature characterization vector, and the third corresponding relationship; Obtaining a fourth corresponding relationship between the feature representation vector, the feature dimension number and the fusion quality index; The fusion quality indicator is determined based on the first feature representation vector, the second feature representation vector, the number of feature dimensions and the fourth corresponding relationship.

6. The method according to claim 3, characterized in that The first branch network further includes a prediction layer connected to the projection head, and the method further includes: When the comprehensive evaluation index is less than the evaluation index threshold, inputting the first feature representation vector into the prediction layer of the first branch network to obtain first prediction data; Based on the first prediction data, symmetrically determine second prediction data corresponding to the second feature characterization vector; Calculating a first correlation coefficient between the second prediction data and the first feature characterization vector, and calculating a second correlation coefficient between the first prediction data and the second feature characterization vector based on a gradient stopping strategy of the second branch network; Determining a contrast loss function of the dual-branch network based on the first correlation coefficient, the second correlation coefficient, and a correlation factor; Based on the contrast loss function of the dual-branch network and the gradient stopping strategy of the second branch network, the parameters of the dual-branch network are updated, and the step of inputting the labeled vibration image and the unlabeled vibration image into the dual-branch network for feature extraction is returned to obtain corresponding feature representation vectors respectively.

7. The method according to claim 1, characterized in that The step of fusing the feature representation vectors to obtain fused features comprises: Performing a residual operation on the feature representation vector to obtain a residual feature; The weighted residual features are connected in multiple levels to obtain the initial fusion features; The initial fusion feature is input into a feature fusion classification module to obtain the fusion feature, wherein the feature fusion classification module includes a first convolutional layer, a normalization layer, a first activation layer, a channel attention layer, a pooling layer, a second convolutional layer and a second activation layer connected in sequence.

8. The method according to any one of claims 1 to 7, characterized in that After the step of determining the life prediction result of the gearbox input shaft bearing based on the fusion feature, the method further includes: Based on the life prediction results of the transmission input shaft bearing, maintenance recommendations are determined.

9. A gearbox input shaft bearing life prediction device, characterized in that: The device comprises: A feature extraction module is used to obtain vibration signals of the gearbox input shaft bearing at different times during the entire life cycle, convert the segmented vibration signals into corresponding vibration images, and extract sample vibration images from the vibration images; The feature extraction module is further used to determine the physical model life index of the gearbox input shaft bearing based on the life prediction physical model of the gearbox input shaft bearing; The feature extraction module is further used to annotate a preset number of sample vibration images based on the physical model life index to obtain annotated vibration images and unannotated vibration images; The feature extraction module is further used to input the annotated vibration image and the unannotated vibration image into a dual-branch network for feature extraction, and obtain corresponding feature representation vectors respectively; A feature fusion module, used to calculate the evaluation index of the feature representation vector, and determine the comprehensive evaluation index based on the evaluation index, wherein the evaluation index at least includes a feature consistency index, a separability index, a robustness index, an overall feature consistency index and a fusion quality index; The feature fusion module is further used to fuse the feature representation vectors to obtain a fusion feature when the comprehensive evaluation index is greater than or equal to the evaluation index threshold; The life prediction module is used to determine the life prediction result of the gearbox input shaft bearing based on the fusion feature.

10. A gearbox input shaft bearing life prediction device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the method for predicting the life of a gearbox input shaft bearing according to any one of claims 1 to 8.

Images