Fault diagnosis method and device for rolling bearing and maintenance platform
By converting the one-dimensional time series signal into two-dimensional images and using the ViL-YOLOv8 model combined with ACmix and MPDIoU loss function, the problem of insufficient generalization performance of rolling bearing fault diagnosis method is solved, and more accurate and robust fault identification is achieved.
Patent Information
- Application Number
- CN202411400238.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-09
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing rolling bearing fault diagnosis methods have insufficient generalization performance and cannot effectively deal with the multi-type and multi-scale characteristics of texture analysis, resulting in poor noise resistance and poor generalization performance.
MTF is used to convert one-dimensional time series signals into two-dimensional images, and ViL is used as the Backbone of the YOLOv8 model, combining ACmix and MPDIoU loss functions to extract the grayscale image of fault characteristics to achieve fault diagnosis.
It improves the generalization performance of rolling bearing fault diagnosis, eliminates interference from changing textures, realizes accurate fault diagnosis, and enhances the model's feature fusion and noise resistance.
Smart Images

Figure CN118913697A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a rolling bearing fault diagnosis method, device, computer program product and maintenance platform. Background Art
[0002] Traditional intelligent fault diagnosis methods extract signal features and identify faults by analyzing the time and frequency domain of the original vibration signal. When faced with actual working conditions with rich and varied fault information, this often leads to unsatisfactory diagnostic results and poor generalization performance. Therefore, it is necessary to develop a fault diagnosis method with strong adaptability and less impact from artificially extracted features to address the shortcomings of traditional methods. Compared with the shallow network structure of machine learning models, deep learning models have deeper network layers, richer functions, and stronger adaptive learning feature capabilities. They can integrate feature extraction and feature classification to achieve an end-to-end process from data to diagnostic results. However, the current methods used for bearing fault diagnosis are based on feature analysis based on fixed texture feature types, without considering the multi-type and multi-scale characteristics of texture analysis, resulting in problems such as poor noise resistance and insufficient generalization performance in bearing fault diagnosis methods based on deep learning. Summary of the invention
[0003] The main purpose of the present application is to provide a rolling bearing fault diagnosis method, device, computer program product and maintenance platform to at least solve the problem of insufficient generalization performance of the rolling bearing fault diagnosis method in the prior art.
[0004] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a rolling bearing fault diagnosis method is provided, characterized in that it includes: obtaining multiple vibration signals and corresponding fault conditions, and converting the vibration signals into one-dimensional time series signals, the fault conditions including inner ring fault of the rolling bearing, outer ring fault of the rolling bearing, rolling element fault of the rolling bearing and no fault; using MTF to convert the multiple one-dimensional time series signals into two-dimensional images to obtain multiple fault feature grayscale images; training a first optimization model according to the multiple fault feature grayscale images and the corresponding fault conditions to obtain a fault diagnosis model, the first optimization model is a YOLOv8 model using ViL as Backbone; converting the vibration signal of the rolling bearing to be detected into a two-dimensional image to obtain a target fault feature grayscale image, and inputting the target fault feature grayscale image into the fault diagnosis model to obtain the fault condition of the rolling bearing to be detected.
[0005] Optionally, a first optimization model is trained according to a plurality of the fault feature grayscale images and the corresponding fault conditions to obtain a fault diagnosis model, including: determining the ViL as the YOLOv8 model of the Backbone as the first optimization model; introducing Neck of the first optimization model into ACmix to obtain a second optimization model; and training the second optimization model according to a plurality of the fault feature grayscale images and the corresponding fault conditions to obtain the fault diagnosis model.
[0006] Optionally, a first optimization model is trained according to a plurality of the fault feature grayscale images and the corresponding fault conditions to obtain a fault diagnosis model, including: determining the ViL as the YOLOv8 model of the Backbone as the first optimization model; replacing the loss function of the first optimization model with MPDIoU to obtain a third optimization model; training the third optimization model according to a plurality of the fault feature grayscale images and the corresponding fault conditions to obtain the fault diagnosis model.
[0007] Optionally, a first optimization model is trained according to a plurality of the fault feature grayscale images and the corresponding fault conditions to obtain a fault diagnosis model, including: determining the ViL as the YOLOv8 model of the Backbone as the first optimization model; introducing Neck of the first optimization model into ACmix to obtain a second optimization model; replacing the loss function of the second optimization model with MPDIoU to obtain a fourth optimization model; training the fourth optimization model according to a plurality of the fault feature grayscale images and the corresponding fault conditions to obtain the fault diagnosis model.
[0008] Optionally, the fourth optimization model is trained according to a plurality of the fault feature grayscale images and the corresponding fault conditions to obtain the fault diagnosis model, including: a training step, inputting a target fault feature grayscale image into the fourth optimization model to obtain at least one prediction box, wherein the target fault feature grayscale image is any one of the fault feature grayscale images; an acquisition step, acquiring a bounding box corresponding to the fault condition existing in the target fault feature grayscale image to obtain a GT box; a verification step, using The loss values of the prediction box and the GT box are calculated, where: is the area of the intersection of the prediction box and the GT box, is the area of the set of the prediction box and the GT box, d1 is the distance between the upper left corner of the prediction box and the upper left corner of the GT box, and d2 is the distance between the upper right corner of the prediction box and the upper right corner of the GT box; a determination step, when the loss value is less than the loss threshold, determines that the fault diagnosis of the target fault feature grayscale image is accurate; repeats the training step, the acquisition step, the verification step and the determination step at least once in sequence until the fault diagnosis of all the fault feature grayscale images is completed, and when the proportion of the fault feature grayscale images with accurate fault diagnosis is greater than a predetermined threshold, determines that the current fourth optimization model is the fault diagnosis model.
[0009] Optionally, the target fault feature grayscale image is input into the fault diagnosis model to obtain the fault condition of the rolling bearing to be detected, including: inputting the target fault feature grayscale image into the fault diagnosis model to obtain multiple bounding boxes and corresponding fault condition probabilities, wherein the fault condition probability is the probability of belonging to the corresponding fault condition; when any one of the fault condition probabilities is greater than a threshold, determining that the rolling bearing to be detected has the fault condition corresponding to the fault condition probability.
[0010] Optionally, multiple vibration signals and corresponding fault conditions are obtained, and the vibration signals are converted into one-dimensional time series signals, including: obtaining at least one rolling bearing with an inner ring fault, at least one rolling bearing with an outer ring fault, at least one rolling bearing with a rolling element fault, and at least one rolling bearing without a fault; obtaining the vibration signal of at least one rolling bearing with an inner ring fault under different loads, the vibration signal of at least one rolling bearing with an outer ring fault under different loads, the vibration signal of at least one rolling bearing with a rolling element fault under different loads, and the vibration signal of at least one rolling bearing without a fault under different loads to obtain the multiple vibration signals; converting the multiple vibration signals into time series signals and performing normalization processing to obtain the one-dimensional time series signal.
[0011] According to another aspect of the present application, a rolling bearing fault diagnosis device is provided, comprising: an acquisition unit, used to acquire multiple vibration signals and corresponding fault conditions, and convert the vibration signals into one-dimensional time series signals, wherein the fault conditions include inner ring fault of the rolling bearing, outer ring fault of the rolling bearing, rolling element fault of the rolling bearing and no fault; a conversion unit, used to convert the multiple one-dimensional time series signals into two-dimensional images using MTF to obtain multiple fault feature grayscale images; a training unit, used to train a first optimization model according to the multiple fault feature grayscale images and the corresponding fault conditions to obtain a fault diagnosis model, wherein the first optimization model is a YOLOv8 model using ViL as Backbone; a diagnosis unit, used to convert the vibration signal of the rolling bearing to be detected into a two-dimensional image to obtain a target fault feature grayscale image, and input the target fault feature grayscale image into the fault diagnosis model to obtain the fault condition of the rolling bearing to be detected.
[0012] According to another aspect of the present application, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, any one of the methods described above is implemented.
[0013] According to another aspect of the present application, a maintenance platform is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of the described methods.
[0014] By applying the technical solution of the present application, the method converts the one-dimensional time series signal corresponding to the vibration signal into a two-dimensional fault feature grayscale image, and the YOLOv8 model can be used to identify the fault. However, the Backbone of the YOLOv8 model is a neural network that cannot extract the time series information of the fault feature grayscale image. By using ViL as the Backbone of the YOLOv8 model, it can be ensured that the extracted features contain time series information, and accurate diagnosis of rolling bearing faults can be achieved without considering texture information, eliminating the interference of variable textures, improving generalization performance, and solving the problem of insufficient generalization performance of rolling bearing fault diagnosis methods in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 A hardware structure block diagram of a mobile terminal for executing a rolling bearing fault diagnosis method provided in an embodiment of the present application is shown;
[0016] Figure 2 A schematic flow chart of a rolling bearing fault diagnosis method provided according to an embodiment of the present application is shown;
[0017] Figure 3 A schematic diagram of the structure of a Vision LSTM module provided according to an embodiment of the present application is shown;
[0018] Figure 4 A schematic diagram of the structure of an ACmix module provided according to an embodiment of the present application is shown;
[0019] Figure 5 A schematic diagram of a fourth optimization model structure provided according to an embodiment of the present application is shown;
[0020] Figure 6 A schematic diagram of a GT box and a prediction box provided according to an embodiment of the present application is shown;
[0021] Figure 7 A structural block diagram of a rolling bearing fault diagnosis device provided according to an embodiment of the present application is shown.
[0022] The above drawings include the following reference numerals:
[0023] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION
[0024] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0025] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.
[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0027] For the convenience of description, some nouns or terms involved in the embodiments of the present application are explained below:
[0028] MTF: Markov Transition Field (MTF) is a non-parametric model for time series data analysis. It can capture the state transition probability in time series data and express it as a two-dimensional probability field, which can be used to predict and analyze the dynamic behavior in time series data. That is, the Markov transition field can convert a one-dimensional time series into a two-dimensional image for display.
[0029] YOLOv8: A single-stage target detection algorithm that achieves target detection by dividing the entire image into grid cells and predicting multiple bounding boxes and category probabilities in each cell. The overall structure of YOLOv8 consists of three parts: Backbone, Neck, and Head. Backbone is mainly used for feature extraction, Neck is mainly used for feature fusion, and Head uses a decoupled head structure, which is divided into classification and positioning prediction ends, alleviating the conflict between classification and positioning tasks.
[0030] ViL (Vision LSTM): LSTM stands for Long Short-Term Memory Network. ViL is the application of Long Short-Term Memory Network to the visual field, namely Vision LSTM.
[0031] ACmix attention mechanism: A represents attention mechanism, and C represents convolution operation. ACmix is actually a hybrid model based on attention mechanism and convolution operation. The features generated by convolution and attention are fused through summation operation, which enhances the feature fusion capability of the model.
[0032] MPDIoU (Minimum point distance based on IoU): A loss function for bounding box regression that aims to solve the problem that existing loss functions cannot be effectively optimized when the predicted box and the GT box (true value bounding box) have the same aspect ratio but different height and width values.
[0033] As introduced in the background technology, the generalization performance of the rolling bearing fault diagnosis method in the prior art is insufficient. To solve this technical problem, the embodiments of the present application provide a rolling bearing fault diagnosis method, device, computer program product and maintenance platform.
[0034] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.
[0035] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 FIG. 1 is a hardware structure block diagram of a mobile terminal of a rolling bearing fault diagnosis method according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.
[0036] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the rolling bearing fault diagnosis method in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The above-mentioned specific network example may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.
[0037] In this embodiment, a rolling bearing fault diagnosis method running on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0038] Figure 2 FIG. 1 is a flow chart of a rolling bearing fault diagnosis method according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:
[0039] Step S201, acquiring multiple vibration signals and corresponding fault conditions, and converting the vibration signals into one-dimensional time series signals, wherein the fault conditions include inner ring fault of the rolling bearing, outer ring fault of the rolling bearing, rolling element fault of the rolling bearing, and no fault;
[0040] Step S202, using MTF to convert the multiple one-dimensional time series signals into two-dimensional images to obtain multiple fault feature grayscale images;
[0041] Step S203, training a first optimization model according to the plurality of the above-mentioned fault feature grayscale images and the corresponding above-mentioned fault conditions to obtain a fault diagnosis model, wherein the above-mentioned first optimization model is a YOLOv8 model using ViL as a backbone;
[0042] Step S204, converting the vibration signal of the rolling bearing to be detected into a two-dimensional image to obtain a target fault feature grayscale image, and inputting the target fault feature grayscale image into the fault diagnosis model to obtain the fault condition of the rolling bearing to be detected.
[0043] In the above-mentioned rolling bearing fault diagnosis method, the one-dimensional time series signal corresponding to the vibration signal is converted into a two-dimensional fault feature grayscale image, and the YOLOv8 model can be used to identify the fault. However, the Backbone of the YOLOv8 model is a neural network that cannot extract the time series information of the fault feature grayscale image. Using ViL as the Backbone of the YOLOv8 model can ensure that the extracted features contain time series information, thereby achieving accurate diagnosis of the rolling bearing fault without considering texture information, eliminating the interference of variable texture, improving the generalization performance, and solving the problem of insufficient generalization performance of the rolling bearing fault diagnosis method in the prior art.
[0044] In order to improve the feature fusion capability of the model, in an optional implementation, the above step S203 includes:
[0045] Step S2031, determining the ViL as the YOLOv8 model of the Backbone as the first optimization model;
[0046] Step S2032, introducing Neck of the first optimization model into ACmix to obtain a second optimization model;
[0047] Step S2033, training the second optimization model according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain the fault diagnosis model.
[0048] In the above implementation, the ViL is used as the YOLOv8 model of the Backbone to determine the first optimization model, and the Neck of the first optimization model is introduced into ACmix to obtain the second optimization model. In the second optimization model, the structure of ViL is as follows: Figure 3 As shown in the figure, after the extracted feature map is passed into the ViL module, it is processed by advanced linear segmentation (Linear Projection) to divide the feature map into 9 patches, and then a serial number (i.e. 1~9) is added to each patch. Each patch is then passed into the core module LSTM of ViL for processing. These 9 patches can be processed in parallel, and the input is the Vision-LSTM Encoder (Vision-LSTM Encoder) to extract the encoding features, and the input is the prediction head (Linear Head) to obtain the predicted feature map. In other words, in the Vision-LSTM model, the extracted feature map can be regarded as being input into the LSTM at each time step in the form of a sequence for processing. The LSTM network will process the input image feature sequence, and use its time series processing capabilities to understand the dynamic changes and contextual information in the image. Finally, it outputs the feature map processed by LSTM, and then performs subsequent convolution operations. The structure of ACmix is shown as follows Figure 4As shown in the figure, in ACmix, the input features are projected through three 1×1 convolutions and then reshaped into N pieces. At this point, a rich set of intermediate features including 3×N feature maps can be obtained. The intermediate features are then passed to the convolution branch and the self-attention branch for processing, and these intermediate features are reused and aggregated according to different paradigms. This can not only utilize the global perception ability of the self-attention mechanism, but also capture local features through convolution. Convolution branch: Integrate the intermediate features through the fully connected layer (Fully Connection), and then use one pixel in the convolution kernel to generate the entire feature map, and obtain the feature map that only belongs to the convolution kernel pixel. The Shift operator is used to fuse the output of the feature map. Self-attention branch: Collect the intermediate features into N groups, each group contains three features, corresponding to query, key and value. Finally, the features obtained by the convolution branch and the self-attention branch are weighted and fused through hyperparameters to obtain the final feature map.
[0049] In order to prevent effective optimization in special cases, in an optional implementation, the above step S203 includes:
[0050] Step S2034, determining the ViL as the YOLOv8 model of the Backbone as the first optimization model;
[0051] Step S2035, replacing the loss function of the first optimization model with MPDIoU to obtain a third optimization model;
[0052] Step S2036, training the third optimization model according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain the fault diagnosis model.
[0053] In the above implementation, the original binary cross entropy loss function in YOLOv8 is replaced by a bounding box regression loss function based on MPDIoU. After using MDPIoU, the fault diagnosis model can better distinguish the background and foreground in the image to solve the problem that the existing loss function cannot be effectively optimized when the predicted box and the GT box have the same aspect ratio but different height and width values.
[0054] In order to improve the accuracy of fault identification, in an optional implementation manner, the above step S203 includes:
[0055] Step S2037, determining the ViL as the YOLOv8 model of the Backbone as the first optimization model;
[0056] Step S2038, introducing Neck of the first optimization model into ACmix to obtain a second optimization model;
[0057] Step S2039, replacing the loss function of the second optimization model with MPDIoU to obtain a fourth optimization model;
[0058] Step S20310, training the fourth optimization model according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain the fault diagnosis model.
[0059] In the above implementation, the overall structure of the YOLOv8 model includes four parts: Input, Backbone, Neck, and Head. The fourth optimization model (fault diagnosis model based on ViL-ACmix-YOLOv8) obtained by transforming the YOLOv8 model is as follows: Figure 5As shown in the figure, Backbone is mainly used for feature extraction. The ViL module is introduced as the backbone, and its time series processing capability is used to understand the dynamic changes and contextual information in the image, and enhance the model's ability to express time-related features. The Neck part is mainly used for feature fusion. The ACmix module is introduced to replace the two Conv modules in the YOLOv8 model to improve the model's feature aggregation capability. The Head part uses a decoupled head structure, which is divided into classification and positioning prediction ends, alleviating the conflict between classification and positioning tasks. First, the input module loads the image, resizes it (in order to adapt to the input size of the YOLOv8 network, the image will be adjusted to a specific resolution), uses Mosaic processing for data enhancement (data enhancement only occurs in the training phase, such as random cropping, rotation, scaling, color jittering, etc.), and normalizes and other image preprocessing operations. Then, the preprocessed image is passed to the backbone of the model. The image first passes through a series of convolutional layers, which usually include convolution operations, batch normalization, and activation functions (such as ReLU or SiLU) to extract multi-scale features of the image, and the c2f module is used to achieve feature fusion to obtain a feature map with high resolution and rich semantic information. The ViL module (i.e., Vision LSTM) is introduced in the backbone. The feature map extracted by convolution is passed to ViL for further processing, and its time series processing capability is used to understand the dynamic changes and contextual information in the image. For example, the c2f_LSTM module has time series processing capabilities compared to the original c2f module. The extracted information is then passed to the fast spatial pyramid pooling layer SPPF module to capture feature information at different levels in the scene and output three feature maps of different scales. These feature maps will be used as inputs to subsequent networks for detection tasks. After the backbone, the Neck part of the model is connected, and the feature maps are further fused and enhanced through upsampling and feature map splicing operations. The ACmix module is introduced to replace the two Conv modules in the YOLOv8 model to improve the model's feature aggregation capabilities. The last part is the Head, which is responsible for the final bounding box prediction and category prediction.
[0060] In order to improve the optimization capability, in an optional implementation, the above step S20310 includes:
[0061] Step S203101, a training step, inputting a target fault feature grayscale image into the fourth optimization model to obtain at least one prediction frame, wherein the target fault feature grayscale image is any one of the fault feature grayscale images;
[0062] Step S203102, an acquisition step, obtaining a bounding box corresponding to the above fault condition existing in the above target fault feature grayscale image, and obtaining a GT box;
[0063] Step S203103, verification step, using The loss values of the above prediction box and the above GT box are calculated, where: is the area of the intersection of the above prediction box and the above GT box, is the area of the set of the prediction box and the GT box, d1 is the distance between the upper left corner of the prediction box and the upper left corner of the GT box, and d2 is the distance between the upper right corner of the prediction box and the upper right corner of the GT box;
[0064] Step S203104, determining step, in the case where the loss value is less than the loss threshold, determining that the fault diagnosis of the target fault feature grayscale image is accurate;
[0065] Step S203105, repeat the above-mentioned training step, the above-mentioned acquisition step, the above-mentioned verification step and the above-mentioned determination step in sequence at least once until the fault diagnosis of all the above-mentioned fault feature grayscale images is completed, and when the proportion of the above-mentioned fault feature grayscale images with accurate fault diagnosis is greater than a predetermined threshold, determine that the current above-mentioned fourth optimization model is the above-mentioned fault diagnosis model.
[0066] In the above implementation, if Figure 6 As shown, the image has a width of w and a height of h, (x1 pre ,y1 pre ) and (x2 pre ,y2 pre ) are the coordinates of the upper left corner and the lower right corner of the prediction box, respectively, (x1 gt ,y1 gt ) and (x2 gt ,y2 gt ) are the coordinates of the upper left corner and lower right corner of the GT box, respectively, and d1 represents the upper left corner of the prediction box (x1 pre ,y1 pre ) and the upper left corner of the GT box (x1 gt ,y1 gt ), d2 represents the distance between the lower right corner of the prediction box (x2 pre ,y2 pre ) and the lower right corner of the GT box (x2 gt ,y2 gt ), the calculation formula of the boundary regression loss function based on MPDIoU is: , , , , the above loss function can also be reflected in the loss value when the prediction box and the GT box have the same aspect ratio but different height and width values, avoiding the problem that this situation cannot be effectively optimized.
[0067] In order to determine the fault type, in an optional implementation manner, the above step S204 includes:
[0068] Step S2041, inputting the target fault feature grayscale image into the fault diagnosis model to obtain a plurality of bounding boxes and corresponding fault condition probabilities, wherein the fault condition probabilities are probabilities of belonging to the corresponding fault conditions;
[0069] Step S2042: when the probability of any one of the above fault conditions is greater than a threshold, determine that the above fault condition corresponding to the above fault condition probability exists in the above rolling bearing to be tested.
[0070] In the above implementation, the fault diagnosis model divides the target fault feature grayscale image into grid units, and predicts multiple bounding boxes and fault condition probabilities in each unit. If any one of the above fault condition probabilities is greater than a threshold, it is determined that the above-mentioned rolling bearing to be tested has the above-mentioned fault condition corresponding to the above-mentioned fault condition probability. For example, the fault condition probability corresponding to the bounding box is the probability of an inner ring fault of 80%, which is greater than a threshold of 75%, then it is determined that the above-mentioned rolling bearing to be tested has an inner ring fault. Of course, multiple bounding boxes may have multiple faults.
[0071] In order to obtain training data, in an optional implementation, the above step S201 includes:
[0072] Step S2011, obtaining at least one rolling bearing with an inner ring fault, at least one rolling bearing with an outer ring fault, at least one rolling bearing with a rolling element fault, and at least one rolling bearing without a fault;
[0073] Step S2012, obtaining vibration signals of at least one rolling bearing with an inner ring fault under different loads, vibration signals of at least one rolling bearing with an outer ring fault under different loads, vibration signals of at least one rolling bearing with a rolling element fault under different loads, and vibration signals of at least one rolling bearing without a fault under different loads, to obtain a plurality of the above vibration signals;
[0074] Step S2013, converting the plurality of vibration signals into time series signals and performing normalization processing to obtain the one-dimensional time series signals.
[0075] In the above implementation, the collected original vibration signal is subjected to data preprocessing, and a plurality of whole one-dimensional time series signals obtained at different fault locations (inner ring fault, outer ring fault, rolling element fault) under different loads are normalized to obtain a plurality of one-dimensional time series signals representing different fault characteristic information; in addition, after the one-dimensional time series signal, the collected one-dimensional time series signal is converted into a two-dimensional image using MTF, and the grayscale images of the same fault characteristics are integrated into a data set, which includes four types of data samples, normal samples, with category 0, inner ring fault samples, with category 1, outer ring fault samples, with category 2, and rolling element fault samples, with category 3; the data set constructed in step 2 is divided according to a ratio of 7:3 to obtain a training set and a test set that meet the requirements of the diagnostic model; the constructed training data set is used to train the fault diagnosis model.
[0076] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0077] The embodiment of the present application also provides a rolling bearing fault diagnosis device. It should be noted that the rolling bearing fault diagnosis device of the embodiment of the present application can be used to execute the rolling bearing fault diagnosis method provided in the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.
[0078] The rolling bearing fault diagnosis device provided in the embodiment of the present application is introduced below.
[0079] Figure 7 1 is a structural block diagram of a rolling bearing fault diagnosis device according to an embodiment of the present application. Figure 7 As shown, the device comprises:
[0080] The acquisition unit 10 is used to acquire multiple vibration signals and corresponding fault conditions, and convert the above vibration signals into one-dimensional time series signals, wherein the above fault conditions include inner ring fault of the rolling bearing, outer ring fault of the rolling bearing, rolling element fault of the rolling bearing, and no fault;
[0081] A conversion unit 20, configured to convert the plurality of one-dimensional time series signals into two-dimensional images by using MTF to obtain a plurality of fault feature grayscale images;
[0082] A training unit 30 is used to train a first optimization model according to the plurality of the above-mentioned fault feature grayscale images and the corresponding above-mentioned fault conditions to obtain a fault diagnosis model, wherein the above-mentioned first optimization model is a YOLOv8 model using ViL as a backbone;
[0083] The diagnosis unit 40 is used to convert the vibration signal of the rolling bearing to be detected into a two-dimensional image to obtain a target fault feature grayscale image, and input the above target fault feature grayscale image into the above fault diagnosis model to obtain the fault condition of the above rolling bearing to be detected.
[0084] In the above-mentioned rolling bearing fault diagnosis device, the one-dimensional time series signal corresponding to the vibration signal is converted into a two-dimensional fault feature grayscale image, and the YOLOv8 model can be used to identify the fault. However, the Backbone of the YOLOv8 model is a neural network that cannot extract the time series information of the fault feature grayscale image. Using ViL as the Backbone of the YOLOv8 model can ensure that the extracted features contain time series information, thereby achieving accurate diagnosis of the rolling bearing fault without considering texture information, eliminating the interference of variable texture, improving the generalization performance, and solving the problem of insufficient generalization performance of the rolling bearing fault diagnosis method in the prior art.
[0085] In order to improve the feature fusion capability of the model, in an optional implementation, the training unit includes:
[0086] A first determination module is used to determine the ViL as the YOLOv8 model of the Backbone as the first optimization model;
[0087] A first optimization module, used for introducing Neck of the first optimization model into ACmix to obtain a second optimization model;
[0088] The first training module is used to train the second optimization model according to the plurality of the above-mentioned fault feature grayscale images and the corresponding above-mentioned fault conditions to obtain the above-mentioned fault diagnosis model.
[0089] In the above implementation, the ViL is used as the YOLOv8 model of the Backbone to determine the first optimization model, and the Neck of the first optimization model is introduced into ACmix to obtain the second optimization model. In the second optimization model, the structure of ViL is as follows: Figure 3As shown in the figure, after the extracted feature map is passed into the ViL module, it is processed by advanced linear segmentation (Linear Projection) to divide the feature map into 9 patches, and then a serial number (i.e. 1~9) is added to each patch. Each patch is then passed into the core module LSTM of ViL for processing. These 9 patches can be processed in parallel, and the input is the Vision-LSTM Encoder (Vision-LSTM Encoder) to extract the encoding features, and the input is the prediction head (Linear Head) to obtain the predicted feature map. In other words, in the Vision-LSTM model, the extracted feature map can be regarded as being input into the LSTM at each time step in the form of a sequence for processing. The LSTM network will process the input image feature sequence, and use its time series processing capabilities to understand the dynamic changes and contextual information in the image. Finally, it outputs the feature map processed by LSTM, and then performs subsequent convolution operations. The structure of ACmix is shown as follows Figure 4 As shown in the figure, in ACmix, the input features are projected through three 1×1 convolutions and then reshaped into N pieces. At this point, a rich set of intermediate features including 3×N feature maps can be obtained. The intermediate features are then passed to the convolution branch and the self-attention branch for processing, and these intermediate features are reused and aggregated according to different paradigms. This can not only utilize the global perception ability of the self-attention mechanism, but also capture local features through convolution. Convolution branch: Integrate the intermediate features through the fully connected layer (Fully Connection), and then use one pixel in the convolution kernel to generate the entire feature map, and obtain the feature map that only belongs to the convolution kernel pixel. The Shift operator is used to fuse the output of the feature map. Self-attention branch: Collect the intermediate features into N groups, each group contains three features, corresponding to query, key and value. Finally, the features obtained by the convolution branch and the self-attention branch are weighted and fused through hyperparameters to obtain the final feature map.
[0090] In order to prevent effective optimization in special cases, in an optional implementation, the training unit includes:
[0091] A second determination module is used to determine the ViL as the YOLOv8 model of the Backbone as the first optimization model;
[0092] A second optimization module is used to replace the loss function of the first optimization model with MPDIoU to obtain a third optimization model;
[0093] The second training module is used to train the third optimization model according to the plurality of grayscale images of the fault characteristics and the corresponding fault conditions to obtain the fault diagnosis model.
[0094] In the above implementation, the original binary cross entropy loss function in YOLOv8 is replaced by a bounding box regression loss function based on MPDIoU. After using MDPIoU, the fault diagnosis model can better distinguish the background and foreground in the image to solve the problem that the existing loss function cannot be effectively optimized when the predicted box and the GT box have the same aspect ratio but different height and width values.
[0095] In order to improve the accuracy of fault identification, in an optional implementation manner, the training unit includes:
[0096] A third determination module is used to determine the ViL as the YOLOv8 model of the Backbone as the first optimization model;
[0097] A third optimization module is used to introduce Neck of the first optimization model into ACmix to obtain a second optimization model;
[0098] A fourth optimization module, used to replace the loss function of the second optimization model with MPDIoU to obtain a fourth optimization model;
[0099] The third training module is used to train the fourth optimization model according to the plurality of the above-mentioned fault feature grayscale images and the corresponding above-mentioned fault conditions to obtain the above-mentioned fault diagnosis model.
[0100] In the above implementation, the overall structure of the YOLOv8 model includes four parts: Input, Backbone, Neck, and Head. The fourth optimization model obtained by transforming the YOLOv8 model is as follows: Figure 5As shown in the figure, Backbone is mainly used for feature extraction. The ViL module is introduced as the backbone, and its time series processing capability is used to understand the dynamic changes and contextual information in the image, and enhance the model's ability to express time-related features. The Neck part is mainly used for feature fusion. The ACmix module is introduced to replace the two Conv modules in the YOLOv8 model to improve the model's feature aggregation capability. The Head part uses a decoupled head structure, which is divided into classification and positioning prediction ends, alleviating the conflict between classification and positioning tasks. First, the input module loads the image, resizes it (in order to adapt to the input size of the YOLOv8 network, the image will be adjusted to a specific resolution), uses Mosaic processing for data enhancement (data enhancement only occurs in the training phase, such as random cropping, rotation, scaling, color jittering, etc.), and normalizes and other image preprocessing operations. Then, the preprocessed image is passed to the backbone of the model. The image first passes through a series of convolutional layers, which usually include convolution operations, batch normalization (BatchNormalization) and activation functions (such as ReLU or SiLU) to extract multi-scale features of the image, and achieves feature fusion through the c2f module to obtain a feature map with high resolution and rich semantic information. The ViL module (i.e., VisionLSTM) is introduced in the backbone. The feature map extracted by convolution is passed to ViL for further processing, and its time series processing capability is used to understand the dynamic changes and contextual information in the image. For example, the c2f_LSTM module has time series processing capabilities compared to the original c2f module. The extracted information is then passed to the fast spatial pyramid pooling layer SPPF module to capture feature information at different levels in the scene and output three feature maps of different scales. These feature maps will be used as inputs to subsequent networks for detection tasks. After the backbone, the Neck part of the model is connected, and the feature maps are further fused and enhanced through upsampling and feature map splicing operations. The ACmix module is introduced to replace the two Conv modules in the YOLOv8 model to improve the model's feature aggregation capabilities. The last part is the Head, which is responsible for the final bounding box prediction and category prediction.
[0101] In order to improve the optimization capability, in an optional implementation, the third training module includes:
[0102] A training submodule, used for executing a training step, inputting a target fault feature grayscale image into the fourth optimization model to obtain at least one prediction frame, wherein the target fault feature grayscale image is any one of the fault feature grayscale images;
[0103] An acquisition submodule is used to execute the acquisition step, obtain a bounding box corresponding to the above fault condition existing in the above target fault feature grayscale image, and obtain a GT box;
[0104] The verification submodule is used to perform the verification steps using The loss values of the above prediction box and the above GT box are calculated, where: is the area of the intersection of the above prediction box and the above GT box, is the area of the set of the prediction box and the GT box, d1 is the distance between the upper left corner of the prediction box and the upper left corner of the GT box, and d2 is the distance between the upper right corner of the prediction box and the upper right corner of the GT box;
[0105] A determination submodule, used to perform a determination step, and determine that the fault diagnosis of the target fault feature grayscale image is accurate when the loss value is less than the loss threshold;
[0106] The repetition submodule is used to execute the above-mentioned training step, the above-mentioned acquisition step, the above-mentioned verification step and the above-mentioned determination step at least once in sequence until the fault diagnosis of all the above-mentioned fault feature grayscale images is completed, and when the proportion of the above-mentioned fault feature grayscale images with accurate fault diagnosis is greater than a predetermined threshold, determine that the current above-mentioned fourth optimization model is the above-mentioned fault diagnosis model.
[0107] In the above implementation, if Figure 6 As shown, the image has a width of w and a height of h, (x1 pre ,y1 pre ) and (x2 pre ,y2 pre ) are the coordinates of the upper left corner and the lower right corner of the prediction box, respectively, (x1 gt ,y1 gt ) and (x2 gt ,y2 gt ) are the coordinates of the upper left corner and lower right corner of the GT box, respectively, and d1 represents the upper left corner of the prediction box (x1 pre ,y1 pre ) and the upper left corner of the GT box (x1 gt ,y1 gt ), d2 represents the distance between the lower right corner of the prediction box (x2 pre ,y2 pre ) and the lower right corner of the GT box (x2 gt ,y2 gt ), the calculation formula of the boundary regression loss function based on MPDIoU is: , , , , the above loss function can also be reflected in the loss value when the prediction box and the GT box have the same aspect ratio but different height and width values, avoiding the problem that this situation cannot be effectively optimized.
[0108] In order to determine the fault type, in an optional implementation manner, the above-mentioned diagnosis unit includes:
[0109] An analysis module, used to input the target fault feature grayscale image into the fault diagnosis model to obtain a plurality of bounding boxes and corresponding fault condition probabilities, wherein the fault condition probabilities are probabilities of belonging to the corresponding fault conditions;
[0110] The fourth determination module is used to determine that the rolling bearing to be tested has the above-mentioned fault condition corresponding to the above-mentioned fault condition probability when the probability of any of the above-mentioned fault conditions is greater than a threshold.
[0111] In the above implementation, the fault diagnosis model divides the target fault feature grayscale image into grid units, and predicts multiple bounding boxes and fault condition probabilities in each unit. If any one of the above fault condition probabilities is greater than a threshold, it is determined that the above-mentioned rolling bearing to be tested has the above-mentioned fault condition corresponding to the above-mentioned fault condition probability. For example, the fault condition probability corresponding to the bounding box is the probability of an inner ring fault of 80%, which is greater than a threshold of 75%, then it is determined that the above-mentioned rolling bearing to be tested has an inner ring fault. Of course, multiple bounding boxes may have multiple faults.
[0112] In order to obtain training data, in an optional implementation, the acquisition unit includes:
[0113] A first acquisition module, used to acquire at least one rolling bearing with an inner ring fault, at least one rolling bearing with an outer ring fault, at least one rolling bearing with a rolling element fault, and at least one rolling bearing without a fault;
[0114] A second acquisition module is used to acquire vibration signals of at least one rolling bearing with an inner ring fault under different loads, vibration signals of at least one rolling bearing with an outer ring fault under different loads, vibration signals of at least one rolling bearing with a rolling element fault under different loads, and vibration signals of at least one rolling bearing without a fault under different loads, to obtain a plurality of the above vibration signals;
[0115] The processing module is used to convert the multiple vibration signals into time series signals and perform normalization processing to obtain the one-dimensional time series signals.
[0116] In the above implementation, the collected original vibration signal is subjected to data preprocessing, and a plurality of whole one-dimensional time series signals obtained at different fault locations (inner ring fault, outer ring fault, rolling element fault) under different loads are normalized to obtain a plurality of one-dimensional time series signals representing different fault characteristic information; in addition, after the one-dimensional time series signal, the collected one-dimensional time series signal is converted into a two-dimensional image using MTF, and the grayscale images of the same fault characteristics are integrated into a data set, which includes four types of data samples, normal samples, with category 0, inner ring fault samples, with category 1, outer ring fault samples, with category 2, and rolling element fault samples, with category 3; the data set constructed in step 2 is divided according to a ratio of 7:3 to obtain a training set and a test set that meet the requirements of the diagnostic model; the constructed training data set is used to train the fault diagnosis model.
[0117] The rolling bearing fault diagnosis device comprises a processor and a memory, wherein the acquisition unit, conversion unit, training unit and diagnosis unit are stored in the memory as program units, and the processor executes the program units stored in the memory to realize corresponding functions. The modules are all located in the same processor; or, the modules are located in different processors in any combination.
[0118] The processor includes a kernel, and the kernel retrieves the corresponding program unit from the memory. One or more kernels can be provided, and the problem of insufficient generalization performance of the rolling bearing fault diagnosis method in the prior art is solved by adjusting kernel parameters.
[0119] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.
[0120] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the rolling bearing fault diagnosis method.
[0121] Specifically, the rolling bearing fault diagnosis method includes:
[0122] Step S201, acquiring multiple vibration signals and corresponding fault conditions, and converting the vibration signals into one-dimensional time series signals, wherein the fault conditions include inner ring fault of the rolling bearing, outer ring fault of the rolling bearing, rolling element fault of the rolling bearing, and no fault;
[0123] Step S202, using MTF to convert the multiple one-dimensional time series signals into two-dimensional images to obtain multiple fault feature grayscale images;
[0124] Step S203, training a first optimization model according to the plurality of the above-mentioned fault feature grayscale images and the corresponding above-mentioned fault conditions to obtain a fault diagnosis model, wherein the above-mentioned first optimization model is a YOLOv8 model using ViL as a backbone;
[0125] Step S204, converting the vibration signal of the rolling bearing to be detected into a two-dimensional image to obtain a target fault feature grayscale image, and inputting the target fault feature grayscale image into the fault diagnosis model to obtain the fault condition of the rolling bearing to be detected.
[0126] An embodiment of the present invention provides a processor, and the processor is used to run a program, wherein the rolling bearing fault diagnosis method is executed when the program is run.
[0127] Specifically, the rolling bearing fault diagnosis method includes:
[0128] Step S201, acquiring multiple vibration signals and corresponding fault conditions, and converting the vibration signals into one-dimensional time series signals, wherein the fault conditions include inner ring fault of the rolling bearing, outer ring fault of the rolling bearing, rolling element fault of the rolling bearing, and no fault;
[0129] Step S202, using MTF to convert the multiple one-dimensional time series signals into two-dimensional images to obtain multiple fault feature grayscale images;
[0130] Step S203, training a first optimization model according to the plurality of the above-mentioned fault feature grayscale images and the corresponding above-mentioned fault conditions to obtain a fault diagnosis model, wherein the above-mentioned first optimization model is a YOLOv8 model using ViL as a backbone;
[0131] Step S204, converting the vibration signal of the rolling bearing to be detected into a two-dimensional image to obtain a target fault feature grayscale image, and inputting the target fault feature grayscale image into the fault diagnosis model to obtain the fault condition of the rolling bearing to be detected.
[0132] An embodiment of the present invention provides a maintenance platform, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, at least the following steps are implemented:
[0133] Step S201, acquiring multiple vibration signals and corresponding fault conditions, and converting the vibration signals into one-dimensional time series signals, wherein the fault conditions include inner ring fault of the rolling bearing, outer ring fault of the rolling bearing, rolling element fault of the rolling bearing, and no fault;
[0134] Step S202, using MTF to convert the multiple one-dimensional time series signals into two-dimensional images to obtain multiple fault feature grayscale images;
[0135] Step S203, training a first optimization model according to the plurality of the above-mentioned fault feature grayscale images and the corresponding above-mentioned fault conditions to obtain a fault diagnosis model, wherein the above-mentioned first optimization model is a YOLOv8 model using ViL as a backbone;
[0136] Step S204, converting the vibration signal of the rolling bearing to be detected into a two-dimensional image to obtain a target fault feature grayscale image, and inputting the target fault feature grayscale image into the fault diagnosis model to obtain the fault condition of the rolling bearing to be detected.
[0137] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program for initializing at least the following method steps:
[0138] Step S201, acquiring multiple vibration signals and corresponding fault conditions, and converting the vibration signals into one-dimensional time series signals, wherein the fault conditions include inner ring fault of the rolling bearing, outer ring fault of the rolling bearing, rolling element fault of the rolling bearing, and no fault;
[0139] Step S202, using MTF to convert the multiple one-dimensional time series signals into two-dimensional images to obtain multiple fault feature grayscale images;
[0140] Step S203, training a first optimization model according to the plurality of the above-mentioned fault feature grayscale images and the corresponding above-mentioned fault conditions to obtain a fault diagnosis model, wherein the above-mentioned first optimization model is a YOLOv8 model using ViL as a backbone;
[0141] Step S204, converting the vibration signal of the rolling bearing to be detected into a two-dimensional image to obtain a target fault feature grayscale image, and inputting the target fault feature grayscale image into the fault diagnosis model to obtain the fault condition of the rolling bearing to be detected.
[0142] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.
[0143] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0144] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0145] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0146] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0147] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0148] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.
[0149] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0150] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0151] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:
[0152] 1) In the rolling bearing fault diagnosis method of the present application, the one-dimensional time series signal corresponding to the vibration signal is converted into a two-dimensional fault feature grayscale image, and the YOLOv8 model can be used to identify the fault. However, the Backbone of the YOLOv8 model is a neural network that cannot extract the time series information of the fault feature grayscale image. By using ViL as the Backbone of the YOLOv8 model, it can be ensured that the extracted features contain time series information, and accurate diagnosis of the rolling bearing fault can be achieved without considering texture information, eliminating the interference of variable texture, improving the generalization performance, and solving the problem of insufficient generalization performance of the rolling bearing fault diagnosis method in the prior art.
[0153] 2) In the rolling bearing fault diagnosis device of the present application, the one-dimensional time series signal corresponding to the vibration signal is converted into a two-dimensional fault feature grayscale image, and the YOLOv8 model can be used to identify the fault. However, the Backbone of the YOLOv8 model is a neural network that cannot extract the time series information of the fault feature grayscale image. Using ViL as the Backbone of the YOLOv8 model can ensure that the extracted features contain time series information, thereby achieving accurate diagnosis of the rolling bearing fault without considering texture information, eliminating the interference of variable texture, improving the generalization performance, and solving the problem of insufficient generalization performance of the rolling bearing fault diagnosis method in the prior art.
[0154] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims
1. A rolling bearing fault diagnosis method, characterized in that: include: Acquire multiple vibration signals and corresponding fault conditions, and convert the vibration signals into one-dimensional time series signals, wherein the fault conditions include inner ring fault of the rolling bearing, outer ring fault of the rolling bearing, rolling element fault of the rolling bearing, and no fault; Using MTF to convert the multiple one-dimensional time series signals into two-dimensional images to obtain multiple fault feature grayscale images; Training a first optimization model according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain a fault diagnosis model, wherein the first optimization model is a YOLOv8 model using ViL as a backbone; The vibration signal of the rolling bearing to be detected is converted into a two-dimensional image to obtain a target fault feature grayscale image, and the target fault feature grayscale image is input into the fault diagnosis model to obtain the fault condition of the rolling bearing to be detected.
2. The method according to claim 1, characterized in that Training a first optimization model according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain a fault diagnosis model includes: Determine the ViL as the YOLOv8 model of the Backbone as the first optimization model; Introducing Neck of the first optimization model into ACmix to obtain a second optimization model; The second optimization model is trained according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain the fault diagnosis model.
3. The method according to claim 1, characterized in that Training a first optimization model according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain a fault diagnosis model includes: Determine the ViL as the YOLOv8 model of the Backbone as the first optimization model; Replacing the loss function of the first optimization model with MPDIoU to obtain a third optimization model; The third optimization model is trained according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain the fault diagnosis model.
4. The method according to claim 1, characterized in that Training a first optimization model according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain a fault diagnosis model includes: Determine the ViL as the YOLOv8 model of the Backbone as the first optimization model; Introducing Neck of the first optimization model into ACmix to obtain a second optimization model; Replacing the loss function of the second optimization model with MPDIoU to obtain a fourth optimization model; The fourth optimization model is trained according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain the fault diagnosis model.
5. The method according to claim 4, characterized in that Training the fourth optimization model according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain the fault diagnosis model comprises: A training step, inputting a target fault feature grayscale image into the fourth optimization model to obtain at least one prediction frame, wherein the target fault feature grayscale image is any one of the fault feature grayscale images; An acquisition step of acquiring a bounding box corresponding to the fault condition existing in the target fault feature grayscale image to obtain a GT box; Verification steps, using The loss values of the prediction box and the GT box are calculated, where: is the area of the intersection of the prediction box and the GT box, is the area of the set of the prediction box and the GT box, d1 is the distance between the upper left corner of the prediction box and the upper left corner of the GT box, and d2 is the distance between the upper right corner of the prediction box and the upper right corner of the GT box; A determination step, in which, when the loss value is less than a loss threshold, the fault diagnosis of the target fault feature grayscale image is determined to be accurate; Repeat the training step, the acquisition step, the verification step and the determination step at least once in sequence until the fault diagnosis of all the fault feature grayscale images is completed, and when the proportion of the fault feature grayscale images with accurate fault diagnosis is greater than a predetermined threshold, determine that the current fourth optimization model is the fault diagnosis model.
6. The method according to any one of claims 1 to 5, characterized in that Inputting the target fault feature grayscale image into the fault diagnosis model to obtain the fault condition of the rolling bearing to be detected includes: Inputting the target fault feature grayscale image into the fault diagnosis model to obtain a plurality of bounding boxes and corresponding fault condition probabilities, wherein the fault condition probability is the probability of belonging to the corresponding fault condition; In the case where any one of the fault condition probabilities is greater than a threshold, it is determined that the rolling bearing to be inspected has the fault condition corresponding to the fault condition probability.
7. The method according to any one of claims 1 to 5, characterized in that Acquiring multiple vibration signals and corresponding fault conditions, and converting the vibration signals into one-dimensional time series signals, including: Acquire at least one rolling bearing with an inner ring fault, at least one rolling bearing with an outer ring fault, at least one rolling bearing with a rolling element fault, and at least one rolling bearing without a fault; Acquire vibration signals of at least one rolling bearing with an inner ring fault under different loads, vibration signals of at least one rolling bearing with an outer ring fault under different loads, vibration signals of at least one rolling element fault under different loads, and vibration signals of at least one rolling bearing without a fault under different loads, to obtain a plurality of vibration signals; The plurality of vibration signals are converted into time series signals and normalized to obtain the one-dimensional time series signal.
8. A rolling bearing fault diagnosis device, characterized in that: include: An acquisition unit, used for acquiring a plurality of vibration signals and corresponding fault conditions, and converting the vibration signals into one-dimensional time series signals, wherein the fault conditions include inner ring fault of the rolling bearing, outer ring fault of the rolling bearing, rolling element fault of the rolling bearing, and no fault; A conversion unit, used to convert the plurality of one-dimensional time series signals into two-dimensional images by using MTF to obtain a plurality of fault feature grayscale images; A training unit, used for training a first optimization model according to the plurality of fault feature grayscale images and the corresponding fault conditions to obtain a fault diagnosis model, wherein the first optimization model is a YOLOv8 model using ViL as a backbone; The diagnosis unit is used to convert the vibration signal of the rolling bearing to be detected into a two-dimensional image to obtain a target fault feature grayscale image, and input the target fault feature grayscale image into the fault diagnosis model to obtain the fault condition of the rolling bearing to be detected.
9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.
10. A maintenance platform, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of claims 1 to 7.
Citation Information
Patent Citations
Bearing fault identification method based on vibration grayscale image and convolutional neural network
CN110595780A
Fault diagnosis model training method and system and diagnosis method and system thereof
CN113780412A
Rolling bearing fault diagnosis method based on GAF-DRSN
CN114595730A
Rolling bearing fault intelligent diagnosis method
CN115290326A
Multi-scale accompanying robot target detection method and system and electronic equipment
CN117523168A
Cited By
Data enhancement method, fault detection model training method and device, and medium
CN121479319A
Data augmentation method, fault detection model training method, device and medium
CN121479319B
