Bearing fault detection method and device and train

Abstract

The invention provides a bearing fault detection method, and relates to the technical field of rail transit, and the method comprises the steps: obtaining original vibration time sequence data, original strain time sequence data and original displacement time sequence data of a to-be-detected bearing; performing enhancement processing on fault time sequence data in the original vibration data to obtain enhanced vibration time sequence data; determining a filtering parameter based on the reference noise data, and filtering the enhanced vibration time sequence data according to the filtering parameter to obtain initial vibration time sequence data; performing multi-modal data fusion and feature extraction processing on the original strain time sequence data, the original displacement time sequence data and the initial vibration time sequence data to obtain fused time sequence features; and inputting the fused time sequence characteristics into a bearing fault detection model, and outputting a fault detection result for the to-be-detected bearing. The invention further provides a bearing fault detection device and a train.

Description

Technical Field

[0001] This disclosure relates to the field of rail transit technology, and more specifically, to a bearing fault detection method, device, and train. Background Technology

[0002] As a critical component of trains, bearings' operational status directly impacts train safety and reliability. Common methods for diagnosing bearing faults include manual inspections and periodic maintenance. However, these manual methods struggle to accurately detect subtle fault signals in the early stages of a failure, often only revealing problems when they have progressed to a more serious stage. This leads to delayed repairs, increasing the risk of train downtime and maintenance costs. Furthermore, manual inspections are labor-intensive and inefficient. Summary of the Invention

[0003] In view of this, the present disclosure provides a bearing fault detection method, device and train.

[0004] One aspect of this disclosure provides a bearing fault detection method, comprising: acquiring original vibration time-series data, original strain time-series data, and original displacement time-series data of a bearing to be tested, wherein the original vibration time-series data includes fault time-series data characterizing a fault that has occurred in the bearing to be tested; enhancing the fault time-series data in the original time-series vibration data to obtain enhanced vibration time-series data; determining filtering parameters based on reference noise data, and filtering the enhanced vibration time-series data according to the filtering parameters to obtain initial vibration time-series data, wherein the reference noise data includes noise data collected at predetermined locations around the bearing to be tested; performing multimodal data fusion and feature extraction processing on the original strain time-series data, original displacement time-series data, and initial vibration time-series data to obtain fused time-series features; inputting the fused time-series features into a bearing fault detection model, and outputting a fault detection result for the bearing to be tested.

[0005] According to embodiments of this disclosure, the bearing fault detection method further includes: acquiring a reference component training sample set corresponding to a reference component, wherein the reference component includes rotating components other than the bearing to be detected; training an initial reference component fault detection model using the reference component training sample set to obtain a target reference component fault detection model; and updating the weights of the target reference component fault detection model to obtain a bearing fault detection model.

[0006] According to embodiments of this disclosure, updating the weights of a target reference component fault detection model to obtain a bearing fault detection model includes: acquiring a bearing training sample set, which includes fault labeling data; inputting the bearing training sample set into the target reference component fault detection model and outputting a prediction result for the bearing training sample set; based on the error between the fault labeling data and the prediction result, performing an iterative update operation on the initial network weights of a predetermined high-level network in the target reference component fault detection model to determine the updated weights of the predetermined high-level network, thereby obtaining the bearing fault detection model.

[0007] According to embodiments of this disclosure, inputting fused temporal features into a bearing fault detection model and outputting fault detection results for the bearing to be detected includes: dividing the fused temporal features into multiple sets of fused features according to multiple time periods; inputting the multiple sets of fused features into the bearing fault detection model and outputting multiple fault states; and determining the fault detection result based on the multiple time periods corresponding to the multiple sets of fused features and the multiple fault states.

[0008] According to embodiments of this disclosure, the faulty bearing detection method further includes: determining the value of a preset parameter based on the vibration frequency range of the original vibration time sequence data.

[0009] According to embodiments of this disclosure, multimodal data fusion and feature extraction processing are performed on the original strain time series data, original displacement time series data, and initial vibration time series data to obtain fused time series features. This includes: preprocessing the original strain time series data, original displacement time series data, and initial vibration time series data to obtain target strain time series data, target displacement time series data, and target vibration time series data; and using a target adversarial network to perform multimodal data fusion and feature extraction on the target strain time series data, target displacement time series data, and target vibration time series data to obtain fused time series features.

[0010] According to embodiments of this disclosure, the faulty bearing detection method further includes: acquiring a multimodal data training sample set; inputting the multimodal data training sample set into an initial adversarial network; performing an iterative update operation on the initial weights of the generator and discriminator in the initial adversarial network to obtain the target weights of the generator and discriminator respectively; wherein performing a prediction operation includes: determining the loss function of the generator and discriminator respectively; and updating the initial weights according to the loss function.

[0011] According to embodiments of this disclosure, preprocessing the original strain time series data, original displacement time series data, and initial vibration time series data includes: performing time calibration processing on the original strain time series data, original displacement time series data, and initial vibration time series data; and performing normalization processing on the time-calibrated original strain time series data, original displacement time series data, and initial vibration time series data to obtain the target strain signal, target displacement signal, and target vibration signal.

[0012] Another aspect of this disclosure provides a bearing fault detection device, comprising: an acquisition module for acquiring original vibration time-series data, original strain time-series data, and original displacement time-series data of a bearing to be tested, wherein the original vibration time-series data includes fault time-series data characterizing a fault that has occurred in the bearing to be tested; an enhancement module for enhancing the fault time-series data in the original vibration time-series data to obtain enhanced vibration time-series data; a filtering module for determining filtering parameters based on reference noise data and filtering the enhanced vibration time-series data according to the filtering parameters to obtain initial vibration time-series data, wherein the reference noise data includes noise data acquired at predetermined locations around the bearing to be tested; a data fusion and feature extraction module for performing multimodal data fusion and feature extraction processing on the original strain time-series data, original displacement time-series data, and initial vibration time-series data to obtain fused time-series features; and an input module for inputting the fused time-series features into a bearing fault detection model and outputting fault detection results for the bearing to be tested.

[0013] Another aspect of this disclosure provides a train, including: a bearing to be tested; sensors for acquiring raw vibration time-series data, raw strain time-series data, and raw displacement time-series data of the bearing to be tested; and fault detection equipment for performing the steps of the above-described fault bearing detection method. Another aspect of this disclosure provides an electronic device, including:

[0014] One or more processors;

[0015] Memory, used to store one or more programs.

[0016] Specifically, when one or more programs are executed by one or more processors, the one or more processors implement the above method.

[0017] Another aspect of this disclosure provides a computer-readable storage medium storing computer-executable instructions that, when executed, are used to implement the methods described above.

[0018] Another aspect of this disclosure provides a computer program product including computer-executable instructions that, when executed, are used to implement the methods described above. Attached Figure Description

[0019] The above and other objects, features and advantages of this disclosure will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 The illustration schematically shows an exemplary system architecture for which the bearing fault detection method, apparatus, and train of this disclosure can be applied;

[0021] Figure 2 A flowchart illustrating a bearing fault detection method according to an embodiment of the present disclosure is shown schematically.

[0022] Figure 3 A flowchart illustrating a bearing fault detection method according to another embodiment of the present disclosure is shown schematically;

[0023] Figure 4 A block diagram of a bearing fault detection apparatus according to an embodiment of the present disclosure is shown schematically;

[0024] Figure 5 A block diagram of a train according to an embodiment of the present disclosure is schematically shown; and

[0025] Figure 6 A block diagram of an electronic device suitable for implementing a bearing fault detection method according to an embodiment of the present disclosure is shown schematically. Detailed Implementation

[0026] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0029] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0030] In the embodiments disclosed herein, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of data (e.g., including but not limited to user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures have been taken to prevent unauthorized access to user personal information data and to safeguard user personal information security, network security, and national security.

[0031] In the embodiments disclosed herein, user authorization or consent is obtained before acquiring or collecting user personal information.

[0032] On the one hand, bearing failures typically don't cause obvious damage until the middle or late stages. In the early stages of a failure, the bearing usually only exhibits minor anomalies, such as localized micro-wear (e.g., tiny wear pits on the inner or outer raceway surface), fine cracks on the bearing surface, tiny metal spalling points, and minor deformation of the cage. However, because the degree of the anomaly is extremely minor, the fault signal strength is far lower than the background interference during normal bearing operation (background interference includes track excitation, motor noise, etc.), meaning the fault signal strength is relatively weak.

[0033] Because the vibration signal generated by early failures is extremely weak and much lower than the background noise, and the vibration signal generated by early failures is mixed with the vibration signal of normal bearing operation (such as the basic vibration of the rolling elements rolling smoothly), environmental noise (such as the friction noise between the wheel and the track, the wind resistance noise of the train, etc.), the vibration signal generated by early failures is difficult to detect manually.

[0034] On the other hand, with the rapid development of rail transit systems and the continuous increase in operating mileage, the number of bearings is enormous, resulting in a huge workload for manual inspection and extremely low efficiency, thus failing to meet the needs of efficient operation and maintenance of modern rail transit.

[0035] In view of this, embodiments of the present disclosure provide a bearing fault detection method, comprising: acquiring original vibration time-series data, original strain time-series data, and original displacement time-series data of the bearing to be tested, wherein the original vibration time-series data includes fault time-series data characterizing that the bearing to be tested has failed; enhancing the fault time-series data in the original vibration time-series data to obtain enhanced vibration time-series data; determining filtering parameters based on reference noise data, and filtering the enhanced vibration time-series data according to the filtering parameters to obtain initial vibration time-series data, wherein the reference noise data includes noise data collected at predetermined locations around the bearing to be tested; performing multimodal data fusion and feature extraction processing on the original strain time-series data, original displacement time-series data, and initial vibration time-series data to obtain fused time-series features; inputting the fused time-series features into a bearing fault detection model, and outputting fault detection results for the bearing to be tested.

[0036] Figure 1 The illustration schematically depicts an exemplary system architecture for which the bearing fault detection method, apparatus, and train of this disclosure can be applied. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, in order to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments or scenarios.

[0037] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0038] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, and / or social media platform software, etc. (for example only).

[0039] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0040] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process received user requests and other time-series data, and feed back the processing results (such as web pages, information, or time-series data obtained or generated according to user requests) to the terminal devices.

[0041] It should be noted that the bearing fault detection method provided in this embodiment can generally be executed by server 105. Correspondingly, the bearing fault detection system provided in this embodiment can generally be located in server 105. The bearing fault detection method provided in this embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the bearing fault detection system provided in this embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Alternatively, the bearing fault detection method provided in this embodiment can also be executed by the first terminal device 101, the second terminal device 102, or the third terminal device 103, or by other terminal devices different from the first terminal device 101, the second terminal device 102, or the third terminal device 103. Accordingly, the bearing fault detection system provided in this embodiment can also be installed in the first terminal device 101, the second terminal device 102 or the third terminal device 103, or in other terminal devices different from the first terminal device 101, the second terminal device 102 or the third terminal device 103.

[0042] For example, a user can initiate an instruction to obtain the fault detection results of the bearing to be tested through a first terminal device 101, a second terminal device 102, or a third terminal device 103. In response to the above instruction, the server 105 can execute a bearing fault detection method, including: acquiring the original vibration time series data, original strain time series data, and original displacement time series data of the bearing to be tested, wherein the original vibration time series data includes fault time series data characterizing that the bearing to be tested has failed; enhancing the fault time series data in the original vibration time series data to obtain enhanced vibration time series data; determining filtering parameters based on reference noise data, and filtering the enhanced vibration time series data according to the filtering parameters to obtain initial vibration time series data, wherein the reference noise data includes noise data collected at predetermined positions around the bearing to be tested; performing multimodal data fusion and feature extraction processing on the original strain time series data, original displacement time series data, and initial vibration time series data to obtain fused time series features; inputting the fused time series features into the bearing fault detection model, and outputting the fault detection results for the bearing to be tested.

[0043] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0044] Figure 2A flowchart illustrating a bearing fault detection method according to an embodiment of the present disclosure is shown schematically.

[0045] like Figure 2 As shown, the method includes operations S210~S250.

[0046] In operation S210, the original vibration time series data, original strain time series data, and original displacement time series data of the bearing to be tested are acquired. The original vibration time series data includes fault time series data that characterizes the bearing to be tested as having failed.

[0047] For example, vibration sensors, strain sensors, and displacement sensors can be installed at predetermined locations on the bearing to be tested to collect raw vibration time-series data, raw strain time-series data, and raw displacement time-series data of the bearing to be tested.

[0048] It should be noted that there are no restrictions on the types of vibration sensors, strain sensors, and displacement sensors, as long as they can accurately collect the vibration data, strain data, and displacement data of the bearing under test and meet the stability and reliability requirements of the rail transit operating environment. For example, displacement sensors can include micro-displacement sensors based on optical principles, or other types of displacement sensors.

[0049] For example, when a train is running normally, the vibration signal of the bearing under test on the train has a stable pattern, such as a fixed frequency and a stable amplitude. However, when the bearing under test fails (such as pitting of the raceway, peeling of rolling elements, or cage breakage), the vibration intensity, frequency, and waveform of the vibration signal of the bearing under test change. For example, the amplitude increases, fault characteristic frequencies appear (such as the fault frequencies of rolling elements and outer rings), and the waveform changes from an approximately sine wave to an irregular waveform. Therefore, the vibration sensor collects raw vibration data containing the above-mentioned fault characteristics.

[0050] For example, the original vibration data, original strain data, and original displacement data of the bearing under test can be collected in real time to obtain the original vibration time series data, original strain time series data, and original displacement time series data.

[0051] Compared to methods that use only a single type of data for fault detection, using multimodal data such as raw vibration data, raw strain data, and raw displacement data for fault detection can reduce the risk of misjudgment and missed disks, and improve the accuracy of fault detection.

[0052] In operation S220, the fault time series data in the original vibration time series data is enhanced to obtain enhanced vibration time series data.

[0053] For example, strain data reflects the overall or local stress distribution of the bearing under test. However, the stress changes caused by early faults (such as microcracks) are extremely small and can be masked by the static stress of the structure itself or the dynamic stress of normal operation, resulting in unclear fault characteristics in the strain data and a lack of clear fault identification. Displacement data characterizes the macroscopic positional changes of the bearing under test, but the small impacts of early faults do not directly cause observable macroscopic displacements. Displacement anomalies only appear when the fault develops to a certain extent (such as when cracks expand and cause structural deformation). Therefore, in the early stages of a fault, displacement data basically has no fault characteristics.

[0054] For example, vibration data is more sensitive to early bearing failures. However, in the early stages of a bearing failure, the fault characteristics in the vibration data are weak and easily masked by normal vibrations or background noise. This makes it difficult to accurately identify fault characteristics using methods such as manual analysis or analysis of the time-domain waveform of the vibration signal.

[0055] Optionally, by enhancing the fault time-series data in the original vibration data, the amplitude of the fault features can be amplified and the signal-to-noise ratio of the fault features can be improved, so that the weak fault features that were originally covered by noise can be revealed.

[0056] Optionally, fault timing data can be enhanced using a chaotic oscillator, such as a Duffing chaotic oscillator. The dynamic equation of the Duffing chaotic oscillator is shown in equation (1).

[0057] (1)

[0058] Wherein, the system's state variables are represented by x and y, and the damping coefficient is represented by... The excitation amplitude is represented by F, the excitation frequency by w, and the time by t.

[0059] The vibration data of the bearing under test in the early stage of the fault is characterized by weak fault features and strong noise. Correlation signal enhancement methods are usually difficult to process the vibration data in the early stage of the fault. For example, linear filtering can only filter out noise in a specific frequency range, but it is difficult to effectively separate the fault signal and noise frequency when they overlap. However, chaotic oscillators can effectively process signals with weak fault features and strong noise. Therefore, the fault time series data in the original vibration data can be effectively enhanced.

[0060] In operation S230, filtering parameters are determined based on reference noise data, and the enhanced vibration time series data is filtered according to the filtering parameters to obtain initial vibration time series data. The reference noise data includes noise data collected at predetermined locations around the bearing to be tested.

[0061] While the weak fault signals in the enhanced vibration time series data are amplified, noise components may still remain. Therefore, the amplified main input signal and the acquired reference noise signal can be input into an adaptive filter to reduce the noise components in the enhanced vibration time series data and improve the signal-to-noise ratio.

[0062] Optionally, the reference noise data includes noise data collected near the bearing to be tested, which has the same frequency distribution and amplitude variation pattern as the noise of the bearing to be tested.

[0063] Optionally, reference noise data and enhanced vibration time-series data can be input into an adaptive filter, and the filtering parameters can include the step size of the adaptive filter. Determining the filtering parameters based on the reference noise data can include updating the step size based on the reference noise data.

[0064] Alternatively, the adaptive filter can use variable step size minimum mean square to cancel noise. The formula for updating the step size can be shown in equation (2).

[0065] (2)

[0066] Wherein, the step size of the nth step is expressed as The step size adjustment factor is expressed as The constant controlling the step size update magnitude is expressed as: The error signal is e(n).

[0067] Optionally, the formula for calculating the error signal can be as shown in equation (3).

[0068] (3)

[0069] The desired signal is represented as d (n) The output signal of the adaptive filter at step n is represented as y(n).

[0070] According to embodiments of this disclosure, by adaptively adjusting the step size, convergence is achieved to cancel out noise components in the signal after the main input has been enhanced by the chaotic oscillator, thereby improving the signal-to-noise ratio.

[0071] Optionally, the filtering method can be adjusted according to actual needs. For example, in addition to adaptive filters, Kalman filters can also be used for filtering. Correspondingly, parameters can be adjusted and optimized according to the characteristics of bearing vibration signals.

[0072] According to embodiments of this disclosure, by installing multiple sensors at key parts of the bearing to be tested to collect multimodal data, using a chaotic oscillator to enhance weak fault signals, and using an adaptive filter to remove noise, the problem that manual fault detection methods are difficult to capture early fault signals of bearings can be solved. This can more effectively capture early fault signals, reduce missed and false alarms, and improve the accuracy of fault detection.

[0073] In operation S240, multimodal data fusion and feature extraction are performed on the original strain time series data, original displacement time series data and initial vibration time series data to obtain fused time series features.

[0074] The original strain time series data can reflect the structural stress changes and load distribution of the bearing under test; the original displacement time series data can reflect the macroscopic deformation and overall motion trend of the bearing under test; the initial vibration time series data can reflect the local impact and high-frequency fault characteristics (such as microcracks, abnormal gear meshing, etc.) of the bearing under test. By performing multimodal data fusion and feature extraction processing on the above different modal data, fused time series features are obtained, which can cover the full-dimensional state of the bearing under test and accurately assess the fault state of the bearing under test.

[0075] When operating S250, the fused timing features are input into the bearing fault detection model, and the fault detection results for the bearing to be detected are output.

[0076] Optionally, by using a bearing fault detection model, the fault type and severity of the bearing to be tested, as well as the fault development trend, can be determined to obtain fault detection results.

[0077] According to embodiments of this disclosure, by acquiring the original vibration time-series data, original strain time-series data, and original displacement time-series data of the bearing to be tested, the fault state of the bearing to be tested can be accurately evaluated from multiple dimensions using multimodal features. By enhancing the fault time-series data in the original vibration data, the amplitude of the fault features in the early stage of the fault can be amplified. Further filtering of the enhanced vibration time-series data obtained by the enhancement process can improve the signal-to-noise ratio, thereby improving the accuracy of subsequent fault detection. Thus, the fault state of the bearing can be accurately detected, and even in the early stage of the fault, accurate detection of the faulty bearing can be achieved.

[0078] According to embodiments of this disclosure, the fault diagnosis method further includes: obtaining a reference component training sample set corresponding to a reference component, wherein the reference component includes rotating components other than the bearing to be tested; training an initial reference component fault detection model using the reference component training sample set to obtain a target reference component fault detection model; and updating the weights of the target reference component fault detection model to obtain a bearing fault detection model.

[0079] Optionally, rotating components include, for example, motor rotors, gears, and ordinary mechanical bearings. The bearing under test operates similarly to other rotating components, rotating at high speed around a fixed axis during operation. Therefore, the fault data of other rotating components will also contain a large number of reusable basic fault characteristics of the bearing under test.

[0080] Since there are few samples of early bearing failures, while training samples of other rotating parts are relatively easy to obtain, the initial reference part fault detection model can be trained using training samples of other rotating parts (such as the reference part training sample set) to obtain the target reference part fault detection model. Then, the target reference part fault detection model can be transferred to the bearing fault detection scenario. Thus, even when there are few samples of early bearing failures, accurate detection of the bearing to be tested can be achieved.

[0081] Optionally, the initial reference component fault detection model may include a deep neural network with a multi-branch structure, for example, it may include two branches, one of which may adopt a convolutional neural network structure based on an attention mechanism, and the other branch may adopt a long short-term memory network.

[0082] For example, for a convolutional neural network (CNN) architecture, the kernel size can be set to 3×3, and the number of parameters can be 32. CNN architectures can be used to process vibration-related features in fused temporal features.

[0083] For example, in a long short-term memory network, the number of hidden layer neurons can be 64, which can be used to process time series features related to strain and displacement in the fusion of temporal features.

[0084] Optionally, the size and number of convolutional kernels can be adjusted according to actual needs, as can the network structure of the deep neural network. For example, gated recurrent units can be used to fuse time-series features related to strain and displacement.

[0085] Optionally, training an initial reference component fault detection model using a reference component training sample set to obtain a target reference component fault detection model may include: pre-training the initial reference component fault detection model using a stochastic gradient descent algorithm combined with a momentum term so that the initial reference component fault detection model can handle the fault characteristics of rotating components.

[0086] According to embodiments of this disclosure, an initial reference component fault detection model is trained using a training sample set of reference components corresponding to the reference component to obtain a target reference component fault detection model. Then, the weights of the target reference component fault detection model are updated to obtain a bearing fault detection model. This can solve the problem of insufficient training samples in the early stage of bearing faults when there are few early samples to be detected. The rotating component provides common features such as harmonics and sidebands for the model to pre-learn, thus solving the problem of insufficient training samples in the early stage of bearing faults.

[0087] According to embodiments of this disclosure, updating the weights of a target reference component fault detection model to obtain a bearing fault detection model includes: acquiring a bearing training sample set, which includes fault labeling data; inputting the bearing training sample set into the target reference component fault detection model and outputting a prediction result for the bearing training sample set; based on the error between the fault labeling data and the prediction result, performing an iterative update operation on the initial network weights of a predetermined high-level network in the target reference component fault detection model to determine the updated weights of the predetermined high-level network, thereby obtaining the bearing fault detection model.

[0088] Optionally, when transferring the trained target reference component fault detection model to the rail transit bearing fault diagnosis task, the target reference component fault detection model can be fine-tuned using the bearing training sample set. Fine-tuning may include inputting the bearing training sample set into the target reference component fault detection model, outputting the prediction results of the bearing training sample set, and performing an iterative update operation on the initial network weights of the predetermined high-level network in the target reference component fault detection model based on the error between the fault labeling data and the prediction results.

[0089] Optionally, the bearing training sample set may include a subset of fault-free bearing samples and a subset of faulty bearing samples. The subset of faulty bearing samples may include, for example, a subset of bearing samples of different early failure types, such as early failure of the inner race, early failure of the outer race, and early failure of the rolling elements. For example, the bearing training sample set may include 500 sets of fault-free bearing samples and 100 sets of samples of each of the different early failure types.

[0090] Optionally, the weights of the lower-level networks in the target reference component fault detection model can be frozen, and only the initial network weights of predetermined higher-level networks can be updated. The lower-level networks in the target reference component fault detection model are used to learn common features among various rotating components (common features include, for example, the fundamental frequency components of vibration, the smoothness of time-series data, etc.), while the higher-level networks are used to learn bearing-specific features.

[0091] By freezing the weights of the underlying network and updating only the initial weights of the predetermined higher-level networks, the model can adapt to the specific characteristics of bearings while preserving general features, reducing the amount of parameter updates and avoiding overfitting under small sample sizes.

[0092] Optionally, performing an update operation may include: determining the parameter gradient based on the error between the fault labeling data and the prediction results; and updating the learning rate according to the parameter gradient to update the initial network weights. The learning rate can be updated as shown in (4).

[0093] (4)

[0094] Where, n t,j Let be the learning rate of the i-th parameter in the t-th iteration, s be the index variable used to iterate from the 1st to the t-th iteration, and g be the learning rate of the i-th parameter. s,i Let be the gradient of the i-th parameter in the s-th iteration. It is a constant.

[0095] For example, it can be based on the gradient g of the parameters at each iteration. s,i Dynamically adjust the learning rate n t,j This allows for weight updates to the initial network weights, completing a fine-tuning process. After multiple fine-tuning iterations, the bearing fault detection model achieves higher accuracy in identifying early-stage faults compared to a model trained from scratch using only the bearing training sample set.

[0096] According to embodiments of this disclosure, by freezing the weights of the lower-level network and only fine-tuning the higher-level network, overfitting can be prevented and the generalization ability of the pre-trained model can be preserved.

[0097] According to embodiments of this disclosure, by constructing a bearing fault detection model including a multi-branch deep neural network, the model is first pre-trained using data from rotating components, then transferred to the rail transit bearing diagnosis task, and the weights of the bottom-level network of the model are frozen, while only the weights of the higher-level network layers are updated. This solves the problem of poor adaptability of the bearing fault detection model to bearing fault modes, making the bearing fault detection model more accurate in identifying early faults on the test set than the model trained from scratch using only rail transit bearing data. At the same time, it can process data and analyze fault evolution in real time during train operation, trigger early warnings in a timely manner, and improve overall operation and maintenance efficiency.

[0098] According to embodiments of this disclosure, inputting fused temporal features into a bearing fault detection model and outputting fault detection results for the bearing to be detected includes: dividing the fused temporal features into multiple sets of fused features according to multiple time periods; inputting the multiple sets of fused features into the bearing fault detection model and outputting multiple fault states; and determining the fault detection result based on the multiple time periods corresponding to the multiple sets of fused features and the multiple fault states.

[0099] Optionally, the original vibration time series data, original strain time series data, and original displacement time series data are acquired in real time, corresponding to multiple acquisition times. Correspondingly, the fused time series features can include fused data corresponding to multiple times. The fused time series features can be divided into multiple sets of fused features based on multiple time periods, with each set of fused features having the same timestamp information. The fault status includes the determination result of whether the bearing under test is faulty based on each set of fused features, and may also include the fault type of the bearing under test.

[0100] Based on multiple fault states and the corresponding time period for each fault state, the changing trend of the fault characteristics of the bearing under test over time can be inferred, and the fault detection result can be obtained. The time period can be characterized by a timestamp. For example, the above-mentioned changing trend can be determined by equation (5).

[0101] (5)

[0102] Among them, time Hidden The maximum probability is expressed as At time t-1, the state is hidden. The maximum probability is expressed as Elements in the state transition probability matrix The elements in are represented as Hidden state The following observations The probability is expressed as .

[0103] According to embodiments of this disclosure, by determining the fault detection result based on multiple time periods corresponding to multiple sets of fused features and multiple fault states, the changing trend of fault features over time can be determined, and the severity of fault development and possible deterioration time points can be predicted.

[0104] Optionally, multi-level early warning thresholds can be set. Fault detection results can be matched with multi-level early warning thresholds to determine the fault handling strategy corresponding to the fault detection results.

[0105] Figure 3 A flowchart illustrating a bearing fault detection method according to another embodiment of the present disclosure is shown schematically.

[0106] like Figure 3 As shown, the method includes operations S301 to S305.

[0107] In operation S301, multimodal data of the bearing under test can be acquired. For example, raw vibration time series data, raw strain time series data, and raw displacement time series data can be acquired.

[0108] In operation S302, fault data is enhanced. For example, fault data in the original vibration time series data can be enhanced.

[0109] In operation S303, the target adversarial network is input and the fused temporal features are output. For example, the above multimodal data can be input into the target adversarial network, which can then perform multimodal data fusion and feature extraction to output fused temporal features.

[0110] In operation S304, the bearing fault detection model is input and the fault detection results are output. For example, fused time-series features can be input into the bearing fault detection model, and the fault detection results can include the fault type of the bearing to be detected.

[0111] When operating S305, different levels of warnings are triggered.

[0112] Optionally, multiple warning thresholds can be set, with different levels of warning thresholds corresponding to different fault handling strategies. For example, if the fault detection results show that fatigue spalling fault characteristics are just beginning to increase, a low-level warning is triggered, and the fault handling strategy includes sending a notification to maintenance personnel so that they begin to focus on the operating status of the bearing and increase the frequency of daily inspections. If the fault detection results reach a pre-set medium-level warning threshold, the fault handling strategy includes generating a prompt message so that maintenance personnel can prepare to replace the relevant spare parts for the bearing and formulate a preliminary maintenance plan. If the fault detection results indicate that the fault is about to seriously affect the safety of train operation, a high-level warning is immediately triggered, and the fault handling strategy includes maintenance personnel arranging a shutdown for inspection and replacement of the corresponding bearing components according to the provided detailed maintenance recommendations.

[0113] According to embodiments of this disclosure, by acquiring multimodal data of the bearing to be tested, enhancing fault signals through enhancement processing, and removing noise by combining adaptive filters, the normalized multimodal data is input into a target adversarial network, and then intelligently diagnosing the bearing to be tested through a bearing fault detection model including a multi-branch deep neural network. This can analyze the fault evolution trend in real time and trigger different levels of early warning, providing strong support for the operation and maintenance of rail transit bearings, effectively improving the accuracy of fault detection and the efficiency of operation and maintenance, and ensuring the safety of train operation.

[0114] According to embodiments of this disclosure, the fault diagnosis method further includes: determining the value of a preset parameter based on the vibration frequency range of the original vibration time series data.

[0115] Optionally, the vibration frequency range may include the characteristic frequency of early bearing failure, such as the inner ring failure frequency. The characteristic frequency of early bearing failure can be estimated by the bearing structure relationship, and the estimation method can be as shown in equation (6).

[0116] (6)

[0117] Among them, f r Where N is the rotational speed of the bearing to be tested, d is the number of rolling elements, d is the diameter of the rolling elements, and D is the pitch circle diameter of the bearing. It represents the contact angle.

[0118] The original vibration time series data of the bearing under test in the early stage of failure has a specific frequency range. Based on the weak vibration frequency range corresponding to the early failure of the bearing, the values ​​of parameters such as damping coefficient, excitation amplitude, and excitation frequency are determined. When the original vibration time series data containing early failure characteristics is input into the chaotic oscillator system, the chaotic oscillator will produce a resonant response because the damping coefficient, excitation amplitude, and excitation frequency are already matched with the fault characteristics. Under the resonance effect, the originally weak fault characteristic components are amplified, thereby enhancing the weak fault characteristics.

[0119] According to embodiments of this disclosure, multimodal data fusion and feature extraction processing are performed on the original strain time series data, original displacement time series data, and initial vibration time series data to obtain fused time series features. This includes: preprocessing the original strain time series data, original displacement time series data, and initial vibration time series data to obtain target strain time series data, target displacement time series data, and target vibration time series data; and using a target adversarial network to perform multimodal data fusion and feature extraction on the target strain time series data, target displacement time series data, and target vibration time series data to obtain fused time series features.

[0120] Optionally, preprocessing can reduce the time error and magnitude difference between the original strain time series data, the original displacement time series data and the initial vibration time series data, making it easier to perform data fusion in the future.

[0121] Optionally, the target adversarial network may include a generator and a discriminator. The generator can be used to convert features from different modalities into fused features of a uniform dimension. The generator may include transposed convolutional layers; for example, four transposed convolutional layers can be set in the generator part. The generator can be used to perform transposed convolution operations, as shown in Equation (7).

[0122] (7)

[0123] in, For the feature map, the activation function is expressed as: From the first The first layer The first channel to the first The kernel weights of the layer are represented as The convolution operation is represented as , No. The layer bias term is represented as , No. The number of channels in a layer is expressed as .

[0124] Optionally, the discriminator section may also be equipped with convolutional layers to perform convolution operations, for example, three convolutional layers may be set. The convolution operation can be as shown in equation (8).

[0125] (8)

[0126] Among them, the discriminator is number 1 The output feature map of the layer is represented as The corresponding convolution kernel weights are expressed as The bias term is represented as , No. The number of channels in a layer is expressed as .

[0127] Optionally, the target strain time series data, target displacement time series data, and target vibration time series data can be input into the target adversarial network. The target adversarial network can fuse the target strain time series data, target displacement time series data, and target vibration time series data, extract features, and output fused time series features.

[0128] According to embodiments of this disclosure, a target adversarial network can be used to uncover deep-seated relationships between target strain time-series data, target displacement time-series data, and target vibration time-series data. Compared to single-modal features, fusion of time-series features can more accurately and comprehensively reflect the state of the bearing under test, thereby improving the accuracy of subsequent fault diagnosis.

[0129] According to embodiments of this disclosure, the fault detection method further includes: acquiring a multimodal data training sample set; inputting the multimodal data training sample set into an initial adversarial network; performing an iterative update operation on the initial weights of the generator and discriminator in the initial adversarial network to obtain the target weights of the generator and discriminator respectively; wherein performing a prediction operation includes: determining the loss function of the generator and discriminator respectively; and updating the initial weights according to the loss function.

[0130] Optionally, the initial adversarial network can be trained using a multimodal data training sample set to obtain the target adversarial network.

[0131] Alternatively, the loss functions for the generator and discriminator can be as shown in equation (9).

[0132] (9)

[0133] Wherein, the input random noise vector of the generator is denoted as z, and the discriminator function is denoted as... The generator function is represented as The expected budget is expressed as x represents the actual fused feature data.

[0134] According to embodiments of this disclosure, the weights are updated based on the loss functions of the generator and discriminator, and after multiple rounds of training, a fused temporal feature that integrates deep-level correlation information of vibration, strain, and displacement multimodal modes is obtained.

[0135] According to embodiments of this disclosure, preprocessing the original strain time series data, original displacement time series data, and initial vibration time series data includes: performing time calibration processing on the original strain time series data, original displacement time series data, and initial vibration time series data; and performing normalization processing on the time-calibrated original strain time series data, original displacement time series data, and initial vibration time series data to obtain the target strain signal, target displacement signal, and target vibration signal.

[0136] Alternatively, the clocks of each sensor node can be calibrated through a network message passing mechanism, so that the data collected by different sensors are strictly aligned in time.

[0137] Optionally, the normalization process may include linear normalization, and the formula for calculating linear normalization may be as shown in equation (10).

[0138] (10)

[0139] The normalized value is expressed as: The original data is represented as The maximum value of this type of data within the collection period is represented as: The minimum value of this type of data within the collection period is represented as: .

[0140] According to embodiments of this disclosure, time calibration can correct time deviations between multimodal data, and normalization can resolve magnitude differences between multimodal data, facilitating subsequent fusion.

[0141] Figure 4 A block diagram of a bearing fault detection apparatus according to an embodiment of the present disclosure is shown schematically.

[0142] like Figure 4 As shown, the bearing fault detection device 400 includes an acquisition module 410, an enhancement module 420, a filtering module 430, a data fusion and feature extraction module 440, and an input module 450.

[0143] The acquisition module 410 is used to acquire the original vibration time-series data, original strain time-series data, and original displacement time-series data of the bearing under test. The original vibration time-series data includes fault time-series data characterizing that the bearing under test has failed. In one embodiment, the acquisition module 410 can be used to perform the operation S210 described above, which will not be repeated here.

[0144] The enhancement module 420 is used to enhance the fault time-series data in the original vibration data to obtain enhanced vibration time-series data. In one embodiment, the enhancement module 420 can be used to perform the operation S220 described above, which will not be repeated here.

[0145] The filtering module 430 is used to determine filtering parameters based on reference noise data, and to filter the enhanced vibration time series data according to the filtering parameters to obtain initial vibration time series data. The reference noise data includes noise data collected at predetermined locations around the bearing to be tested. In one embodiment, the filtering module 430 can be used to perform the operation S230 described above, which will not be repeated here.

[0146] The data fusion and feature extraction module 440 is used to perform multimodal data fusion and feature extraction processing on the original strain time series data, the original displacement time series data, and the initial vibration time series data to obtain fused time series features. In one embodiment, the data fusion and feature extraction module 440 can be used to perform the operation S240 described above, which will not be repeated here.

[0147] The input module 450 is used to input the fused time-series features into the bearing fault detection model and output the fault detection results for the bearing to be detected. In one embodiment, the input module 450 can be used to perform the operation S250 described above, which will not be repeated here.

[0148] According to embodiments of this disclosure, the bearing fault detection device includes an acquisition module, a training module, and a weight update module.

[0149] The acquisition module is used to acquire a training sample set of reference components corresponding to the reference components, wherein the reference components include rotating components other than the bearing to be detected; the training module is used to train an initial reference component fault detection model using the reference component training sample set to obtain a target reference component fault detection model; the weight update module is used to update the weights of the target reference component fault detection model to obtain a bearing fault detection model.

[0150] According to embodiments of this disclosure, the weight update module includes an acquisition submodule, an input submodule, and an iterative update submodule.

[0151] The acquisition submodule is used to acquire bearing training samples, which include fault labeling data; the input submodule inputs the bearing training sample set into the target reference component fault detection model and outputs the prediction results of the bearing training samples; the iterative update submodule is used to perform iterative update operations on the initial network weights of the predetermined high-level network in the target reference component fault detection model based on the error between the fault labeling data and the prediction results, to determine the updated weights of the predetermined high-level network, and obtain the bearing fault detection model.

[0152] According to embodiments of this disclosure, the input module includes a partitioning submodule, an output submodule, and a determination submodule.

[0153] The partitioning submodule is used to divide the fused time-series features into multiple sets of fused features based on multiple time periods; the output submodule is used to input the multiple sets of fused features into the bearing fault detection model and output multiple fault states; the determination submodule is used to determine the fault detection result based on the multiple time periods corresponding to the multiple sets of fused features and the multiple fault states.

[0154] According to embodiments of this disclosure, the bearing fault detection device further includes a determination module.

[0155] The determination module is used to determine the values ​​of preset parameters based on the vibration frequency range of the original vibration time series data.

[0156] According to embodiments of this disclosure, the data fusion and feature extraction module includes a preprocessing submodule, a multimodal data fusion submodule, and a feature extraction submodule.

[0157] The preprocessing submodule is used to preprocess the original strain time series data, original displacement time series data, and initial vibration time series data to obtain target strain time series data, target displacement time series data, and target vibration time series data. The multimodal data fusion and feature extraction submodule is used to perform multimodal data fusion and feature extraction on the target strain time series data, target displacement time series data, and target vibration time series data using a target adversarial network to obtain fused time series features.

[0158] According to embodiments of this disclosure, the bearing fault detection device further includes a sample set acquisition module, a sample set input module, and an initial weight update module.

[0159] The sample set acquisition module includes acquiring a multimodal data training sample set; the sample set input module includes inputting the multimodal data training sample set into the initial adversarial network; the initial weight update module includes performing iterative update operations on the initial weights of the generator and discriminator in the initial adversarial network to obtain the target weights of the generator and discriminator respectively; wherein, performing one round of prediction operation includes: determining the loss function of the generator and discriminator respectively; and updating the initial weights according to the loss function.

[0160] According to embodiments of this disclosure, the preprocessing submodule includes a time calibration unit and a normalization unit.

[0161] The time calibration unit is used to perform time calibration processing on the original strain time series data, the original displacement time series data, and the initial vibration time series data; the normalization unit is used to normalize the original strain time series data, the original displacement time series data, and the initial vibration time series data after time calibration processing to obtain the target strain signal, the target displacement signal, and the target vibration signal.

[0162] Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be implemented by dividing them into multiple modules. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as hardware circuitry, such as a Field-Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a System-on-Chip, a System-on-a-Substrate, a System-on-Package, an Application-Specific Integrated Circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0163] For example, any plurality of the acquisition module 410, enhancement module 420, filtering module 430, data fusion and feature extraction module 440, and input module 450 can be combined into one module / unit / subunit, or any one of these modules / units / subunits can be split into multiple modules / units / subunits. Alternatively, at least part of the functionality of one or more of these modules / units / subunits can be combined with at least part of the functionality of other modules / units / subunits and implemented in one module / unit / subunit. According to embodiments of the present disclosure, at least one of the acquisition module 410, enhancement module 420, filtering module 430, data fusion and feature extraction module 440, and input module 450 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the acquisition module 410, enhancement module 420, filtering module 430, data fusion and feature extraction module 440, and input module 450 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0164] It should be noted that the data processing system part in the embodiments of this disclosure corresponds to the data processing method part in the embodiments of this disclosure. The specific description of the data processing system part is referred to in the data processing method part, and will not be repeated here.

[0165] Figure 5 A block diagram of a train according to an embodiment of the present disclosure is shown schematically.

[0166] like Figure 5 As shown, the train 500 may include a bearing to be tested 510, a sensor 520, and a fault detection device 530.

[0167] According to embodiments of this disclosure, sensor 520 may include, for example, a displacement sensor for real-time acquisition of raw displacement time-series data, or a vibration sensor for real-time acquisition of raw vibration time-series data, wherein the raw vibration time-series data includes fault time-series data characterizing that the bearing under test has failed. The sensor may also include a strain sensor for real-time acquisition of raw strain time-series data.

[0168] The fault detection device 530 can be used to perform the following operations: acquire the original vibration time-series data, original strain time-series data, and original displacement time-series data of the bearing to be tested, wherein the original vibration time-series data includes fault time-series data characterizing that the bearing to be tested has failed; enhance the fault time-series data in the original vibration data to obtain enhanced vibration time-series data; determine the filtering parameters based on the reference noise data, and filter the enhanced vibration time-series data according to the filtering parameters to obtain the initial vibration time-series data, wherein the reference noise data includes noise data collected at predetermined positions around the bearing to be tested; perform multimodal data fusion and feature extraction processing on the original strain time-series data, original displacement time-series data, and initial vibration time-series data to obtain fused time-series features; input the fused time-series features into the bearing fault detection model, and output the fault detection results for the bearing to be tested.

[0169] Figure 6 A block diagram of an electronic device suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0170] like Figure 6 As shown, an electronic device 600 according to an embodiment of this disclosure includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this disclosure.

[0171] RAM 603 stores various programs and timing data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 602 and / or RAM 603. It should be noted that programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in one or more memories.

[0172] According to embodiments of this disclosure, the electronic device 600 may further include an input / output (I / O) interface 606, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0173] According to embodiments of this disclosure, the method flow according to embodiments of this disclosure can be implemented as a computer software program. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611. When the computer program is executed by processor 601, it performs the functions defined in the system of embodiments of this disclosure. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0174] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0175] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0176] For example, according to embodiments of this disclosure, a computer-readable storage medium may include the ROM 602 and / or RAM 603 described above and / or one or more memories other than ROM 602 and RAM 603.

[0177] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this disclosure. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the bearing fault detection method provided in the embodiments of this disclosure.

[0178] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0179] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0180] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on a user's computing device, partially on a user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0181] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features described in the various embodiments of the present disclosure can be combined and / or combined in various ways, even if such combinations are not explicitly described in the present disclosure. In particular, the features described in the various embodiments of this disclosure may be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0182] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A bearing fault detection method, comprising: Acquire the original vibration time series data, original strain time series data, and original displacement time series data of the bearing to be tested. The original vibration time series data includes fault time series data characterizing that the bearing to be tested has failed. The fault time series data in the original vibration time series data is enhanced to obtain enhanced vibration time series data. The filtering parameters are determined based on the reference noise data, and the enhanced vibration time series data is filtered according to the filtering parameters to obtain the initial vibration time series data. The reference noise data includes noise data collected at predetermined positions around the bearing to be tested. Multimodal data fusion and feature extraction are performed on the original strain time series data, the original displacement time series data, and the initial vibration time series data to obtain fused time series features; The fused temporal features are input into the bearing fault detection model, and the fault detection results for the bearing to be detected are output.

2. The method according to claim 1, wherein, The method further includes: Obtain a training sample set of reference components corresponding to the reference components, wherein the reference components include rotating components other than the bearing to be tested; The initial reference component fault detection model is trained using the training sample set of the reference component to obtain the target reference component fault detection model; The target reference component fault detection model is updated with weights to obtain the bearing fault detection model.

3. The method according to claim 2, wherein, The step of updating the weights of the target reference component fault detection model to obtain the bearing fault detection model includes: Obtain a bearing training sample set, which includes fault labeling data; The bearing training sample set is input into the target reference component fault detection model, and the prediction results for the bearing training sample set are output. Based on the error between the fault labeling data and the prediction result, an iterative update operation is performed on the initial network weights of the predetermined high-level network in the target reference component fault detection model to determine the updated weights of the predetermined high-level network, thereby obtaining the bearing fault detection model.

4. The method according to claim 1, wherein inputting the fused time-series features into the bearing fault detection model and outputting the fault detection result for the bearing to be detected includes: The fused temporal features are divided into multiple groups of fused features based on multiple time periods; The multiple sets of fused features are input into the bearing fault detection model to output multiple fault states; The fault detection result is determined based on the multiple time periods corresponding to the multiple sets of fused features and the multiple fault states.

5. The method according to claim 1, wherein, The method further includes: The value of the preset parameter is determined based on the vibration frequency range of the original vibration time series data.

6. The method according to claim 1, wherein, The process of multimodal data fusion and feature extraction of the original strain time series data, the original displacement time series data, and the initial vibration time series data to obtain fused time series features includes: The original strain time series data, the original displacement time series data, and the initial vibration time series data are preprocessed to obtain target strain time series data, target displacement time series data, and target vibration time series data; Using a target adversarial network, multimodal data fusion and feature extraction are performed on the target strain time series data, target displacement time series data, and target vibration time series data to obtain the fused time series features.

7. The method according to claim 6, further comprising: Obtain a multimodal data training sample set; The multimodal data training sample set is input into the initial adversarial network; An iterative update operation is performed on the initial weights of the generator and the discriminator in the initial adversarial network to obtain the target weights of the generator and the discriminator respectively. One round of prediction operations includes: Determine the loss functions for the generator and the discriminator respectively; The initial weights are updated according to the loss function.

8. The method according to claim 6, wherein, The preprocessing of the original strain time series data, the original displacement time series data, and the initial vibration time series data includes: Time calibration processing is performed on the original strain time series data, the original displacement time series data, and the initial vibration time series data; The original strain time series data, the original displacement time series data, and the initial vibration time series data after time calibration are normalized to obtain the target strain signal, the target displacement signal, and the target vibration signal.

9. A bearing fault detection device, comprising: The acquisition module is used to acquire the original vibration time series data, original strain time series data, and original displacement time series data of the bearing to be tested. The original vibration time series data includes fault time series data that characterizes the bearing to be tested as having failed. An enhancement module is used to enhance the fault time-series data in the original vibration time-series data to obtain enhanced vibration time-series data. A filtering module is used to determine filtering parameters based on reference noise data, and to filter the enhanced vibration time series data according to the filtering parameters to obtain initial vibration time series data. The reference noise data includes noise data collected at predetermined positions around the bearing to be tested. The data fusion and feature extraction module is used to perform multimodal data fusion and feature extraction processing on the original strain time series data, the original displacement time series data and the initial vibration time series data to obtain fused time series features; The input module is used to input the fused time-series features into the bearing fault detection model and output the fault detection results for the bearing to be detected.

10. A train, comprising: The bearing to be tested; Sensors are used to collect the original vibration time-series data, original strain time-series data, and original displacement time-series data of the bearing under test. A fault detection device, the fault detection device being used to perform the steps of the method according to any one of claims 1 to 8.

Images