Bearing quality production detection system based on bearing manufacturing

Abstract

The application relates to the technical field of quality inspection systems, in particular to a bearing quality production detection system based on bearing manufacturing, which comprises the following modules: a rotation simulation and signal collection module: responsible for simulating the rotary motion of a bearing and collecting vibration signals under controlled conditions; a signal processing module: enhancing the high-frequency signal part while suppressing the signals of non-target frequency bands; a feature extraction and analysis module: using time-frequency analysis technology to analyze the preprocessed high-frequency vibration signals and extracting high-frequency vibration features related to the characteristics of the high-frequency vibration signals; a multi-source positioning module: based on spatial positioning technology to identify the direction of the vibration source; an adaptive high-frequency vibration feature learning and intelligent source identification module: constructing a graph neural network, and positioning the fault source and the fault type after training. The application provides key information for in-depth understanding of the complex vibration behavior in bearings, and the high-frequency vibration features can reveal subtle bearing abnormalities and improve the accuracy of fault monitoring.

Description

Technical Field

[0001] This invention relates to the field of quality inspection system technology, and in particular to a bearing quality production inspection system based on bearing manufacturing. Background Technology

[0002] Bearings are a critical component of mechanical equipment, and their performance directly affects the stability and reliability of the entire system. In the post-production quality inspection of bearings, multiple factors need to be considered, among which simulating rotating loads and analyzing vibration anomalies is a core element. Traditional methods for detecting bearing vibration anomalies mainly rely on periodic inspections and basic vibration signal analysis. However, these methods often fall short in capturing early signs of failure, especially in identifying and locating complex failure modes.

[0003] Furthermore, existing fault location techniques struggle to pinpoint the exact location of faults, posing challenges to bearing maintenance and repair. The introduction of Direction of Arrival (DOA) estimation offers a new solution to the spatial location problem of vibration sources. By analyzing vibration signals from different directions, DOA estimation can determine the approximate direction of the vibration source, thus providing strong support for spatial fault location. However, this technique still has limitations in accurately identifying fault locations and analyzing fault types.

[0004] Therefore, although bearing fault detection and diagnosis technology has made some progress, there is still room for improvement in the real-time, accurate capture and localization of complex fault modes. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides a bearing quality production inspection system based on bearing manufacturing.

[0006] The bearing quality production inspection system based on bearing manufacturing includes the following modules:

[0007] Rotation simulation and signal acquisition module: responsible for simulating the rotational motion of the bearing and acquiring vibration signals under controlled conditions;

[0008] Signal processing module: Based on bandpass filtering, the captured vibration signal is preprocessed to enhance the high-frequency signal portion while suppressing signals in non-target frequency bands;

[0009] Feature extraction and analysis module: Utilizes time-frequency analysis technology to analyze the preprocessed high-frequency vibration signal, revealing the frequency components of the high-frequency vibration signal that change over time, and extracting high-frequency vibration features related to the characteristics of the high-frequency vibration signal based on the time-frequency analysis results;

[0010] Multi-source localization module: Based on spatial localization technology (direction of arrival (DOA) estimation technology based on sensor array) to identify the direction of vibration source;

[0011] Adaptive high-frequency vibration feature learning and intelligent source identification module: Based on the vibration source direction and high-frequency vibration features identified by spatial positioning technology, a graph neural network is constructed, and after training, the fault source and fault type are located.

[0012] Furthermore, the rotation simulation and signal acquisition module specifically includes a rotation simulation submodule and a signal acquisition submodule;

[0013] The rotating simulation sub-mold body includes:

[0014] Rotary drive unit: The bearing to be inspected is mounted on a rotating shaft, and the rotating shaft is driven to rotate by a motor to simulate the rotational motion of the bearing to be inspected in actual applications;

[0015] Loading condition settings: Simulate the axial or radial load that the bearing experiences during actual operation;

[0016] Speed ​​control unit: Adjusts the rotation speed to cover different working conditions from low speed to high speed;

[0017] The signal acquisition submodule uses a sensor array (such as an accelerometer) to capture the high-frequency vibration signals generated when the bearing is running.

[0018] Furthermore, the signal processing module specifically includes:

[0019] A frequency range is predefined, which matches the frequency of the high-frequency vibration signal expected to be generated when the bearing is running. Using a bandpass filter, only signal components within the predefined frequency range are allowed to pass through, while other signal components below and above this frequency range are suppressed.

[0020] Furthermore, the time-frequency analysis technique, based on short-time Fourier transform, processes the bandpass-filtered high-frequency vibration signal and calculates it as follows: ,in, It is a signal in the time domain. It revolves around the time window function. It is angular frequency;

[0021] Based on the time-frequency analysis results, high-frequency vibration characteristics related to the high-frequency vibration signal properties, including instantaneous energy, are extracted. Spectrum peak High-frequency energy distribution .

[0022] Furthermore, the instantaneous energy extraction calculation includes: within a given time window, integrating the squares of the amplitudes of all frequency components in the time-frequency representation obtained from the STFT, to calculate:

[0023] Instantaneous energy ; indicates that at a given time point t, the instantaneous energy of the signal is obtained by summing the squares of the amplitudes of all frequency components in the short-time Fourier transform (STFT) result at that time point, where, It is the result of the short-time Fourier transform.

[0024] Furthermore, the method for calculating the spectral peak value is to find the point of maximum amplitude within each time window of the STFT result:

[0025] Spectrum peak ,in, Indicates time The frequency corresponding to the peak value of the spectrum. The operation is used to find out why Frequency of reaching the maximum value .

[0026] Furthermore, the high-frequency energy distribution is obtained by calculating the sum of the energy in the high-frequency region of the STFT result and comparing it with the sum of the energy over the entire frequency range, as follows:

[0027] ;

[0028] ;

[0029] ,

[0030] in, It is the sum of energy in the high-frequency region of time. It is in time The sum of energy across the entire frequency range. It is the defined high-frequency region. and These are the minimum and maximum frequencies in STFT analysis. By comparing the ratio of energy in the high-frequency region to the total energy, the proportion and importance of high-frequency components in the signal are quantified. It refers to the high-frequency energy distribution, i.e., the high-frequency ratio.

[0031] Furthermore, the spatial positioning technology in the multi-source positioning module is based on DOA estimation, utilizing the direction of arrival estimation of the sensor array to perform multi-source positioning, and employing the multi-signal classification (MUSIC) algorithm, with the following steps:

[0032] Calculate the covariance matrix of the received signal. ,in, Indicates time The signal vector received by the sensor array, It is the total number of time samples. Indicates conjugate transpose;

[0033] For covariance matrix Perform feature decomposition to obtain the signal subspace and the noise subspace;

[0034] Constructing the MUSIC spectral peak function: ,in, It is a direction vector, which depends on the direction of arrival of the wave. It is the eigenvector matrix of the noise subspace;

[0035] Scan all directions of arrival ,turn up The peak value, the location of which is the estimated direction of the vibration source;

[0036] By analyzing the DOA estimation results of each sensor for the same vibration source, the position coordinates of the vibration source are calculated using the triangulation method.

[0037] Furthermore, the construction of the graph neural network specifically includes:

[0038] Constructing a graph structure: Nodes in the graph structure represent the directions of the identified vibration sources, and the features of the nodes include the corresponding high-frequency vibration characteristics, i.e., instantaneous energy. Spectrum peak High-frequency energy distribution The weights of the edges are defined based on the spatial distance or similarity between the vibration sources;

[0039] Graph Neural Network (GNN) Model Design: Design a GNN model that takes a graph structure as input and learns the complex spatial relationships and characteristic information of vibration sources by aggregating and updating information on graph nodes.

[0040] Model training and validation: The GNN model is trained using the prepared graph structure dataset. Different fault types and their corresponding graph structures are used as inputs, and the fault source location is used as the label. The model parameters are optimized using loss functions and optimization algorithms, and the model performance is evaluated using the validation set.

[0041] Fault source identification and localization: After the model is trained, the vibration source direction and high-frequency vibration characteristics are processed through the same preprocessing and graph construction steps and then input into the trained GNN model. The model will predict the type and specific location of the fault based on the information of the entire graph structure, thereby realizing the comprehensive analysis of multiple vibration sources and the localization of the fault source.

[0042] Furthermore, the graph neural network is specifically represented as follows:

[0043] Graph structure definition: Constructing a graph ,in, It is a set of nodes, each node representing a vibration source, and the node characteristics include high-frequency vibration features. (Instantaneous energy, spectral peak value, and high-frequency energy distribution) and DOA estimation results, It is a set of edges, and the weights of the edges. Calculated based on the spatial distance or similarity between vibration sources;

[0044] Information propagation and aggregation: In each iteration, nodes From its neighboring nodes Represents a node The set of neighboring nodes collects information and updates its own feature representation, using the following calculation formula:

[0045] ,in, It is a node In the Feature representation of the next iteration, UPDATE and These are the update and aggregation functions, which are specifically defined according to different GNN variants, including GCN and GAT;

[0046] Readout layer: For each node (vibration source) in the graph, the readout function READOUT is used to aggregate information from the entire graph, resulting in a graph-level representation used for final fault type and location prediction. ,in, It is the final iteration number. It is a graph-level representation;

[0047] Fault prediction: using graph-level representation Predict the type and location of faults using one or more fully connected layers: ,in, It predicts the type and location of the fault. and These are the weights and biases of the fully connected layer, and softmax is used to normalize the output, making it represent a probability distribution.

[0048] The beneficial effects of this invention are:

[0049] This invention, by combining high-frequency vibration feature analysis with a multi-source positioning module, significantly enhances the ability to identify and distinguish multiple vibration sources in a bearing system. In particular, by comprehensively applying high-frequency vibration features such as instantaneous energy, spectral peak value, and high-frequency energy distribution, it provides key information for a deeper understanding of the complex vibration behavior in bearings. These high-frequency vibration features can reveal subtle bearing anomalies, even in the early stages of development, thereby significantly improving the accuracy of fault diagnosis.

[0050] This invention utilizes a graph neural network (GNN) model to comprehensively consider the spatial relationships between vibration sources and their respective high-frequency vibration characteristics, achieving precise fault location. By learning the spatial distribution and interactions of vibration sources, the GNN model can effectively understand and interpret complex patterns in vibration signals. This deep learning method enables the invention not only to accurately identify the type of fault but also to pinpoint its specific location. For example, if a particular vibration characteristic is more relevant to one side of the bearing, the GNN can use this information to predict which part of the bearing the fault is more likely to be located. Such precise fault location is extremely important for guiding maintenance personnel to perform targeted repairs, significantly reducing the time and labor costs in the diagnosis and repair process, and improving the efficiency and effectiveness of maintenance operations. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a schematic diagram of the system modules according to an embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram of a graph neural network according to an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0055] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0056] like Figure 1-2 As shown, the bearing quality production inspection system based on bearing manufacturing includes the following modules:

[0057] Rotation simulation and signal acquisition module: responsible for simulating the rotational motion of the bearing and acquiring vibration signals under controlled conditions;

[0058] Signal processing module: Based on bandpass filtering, the captured vibration signal is preprocessed to enhance the high-frequency signal portion while suppressing signals in non-target frequency bands;

[0059] Feature extraction and analysis module: Utilizes time-frequency analysis technology to analyze the preprocessed high-frequency vibration signal, revealing the frequency components of the high-frequency vibration signal that change over time, and extracting high-frequency vibration features related to the characteristics of the high-frequency vibration signal based on the time-frequency analysis results;

[0060] Multi-source localization module: Based on spatial localization technology (direction of arrival (DOA) estimation technology based on sensor array) to identify the direction of vibration source;

[0061] Adaptive high-frequency vibration feature learning and intelligent source identification module: Based on the vibration source direction and high-frequency vibration features identified by spatial positioning technology, a graph neural network is constructed, and after training, the fault source and fault type are located.

[0062] The rotation simulation and signal acquisition module specifically includes a rotation simulation submodule and a signal acquisition submodule;

[0063] The rotating simulation sub-mold body includes:

[0064] Rotary drive unit: The bearing to be inspected is mounted on a rotating shaft, and the rotating shaft is driven to rotate by a motor to simulate the rotational motion of the bearing to be inspected in actual applications;

[0065] Loading condition settings: Simulate the axial or radial load that the bearing experiences during actual operation;

[0066] Speed ​​control unit: Adjusts the rotation speed to cover different working conditions from low speed to high speed;

[0067] The signal acquisition submodule employs a sensor array (such as accelerometers) to capture high-frequency vibration signals generated during bearing operation. Specifically, firstly, a group of high-sensitivity accelerometers are evenly distributed and installed at fixed positions around the bearing under test, forming a sensor array. Each accelerometer can independently capture the vibration signals generated by the bearing during operation, especially the high-frequency vibration components. Then, the output signals of all accelerometers are acquired synchronously and recorded in real time. Furthermore, a high-speed data acquisition card and signal processing software are used to perform preliminary synchronization and time stamping of the acquired signals, ensuring the integrity and consistency of the signals in subsequent analysis. Through this method, the signal acquisition submodule can effectively capture the high-frequency vibration signals generated by the bearing under different operating conditions, providing an accurate and reliable data foundation for subsequent signal processing, feature extraction, and fault analysis.

[0068] The signal processing module specifically includes:

[0069] A predefined frequency range is used to match the frequency of the high-frequency vibration signal expected to be generated when the bearing is running. Using a bandpass filter, only signal components within the predefined frequency range are allowed to pass through, while other signal components below and above this frequency range are suppressed. This preprocessing process not only enhances the target high-frequency vibration signal, improving the accuracy and efficiency of subsequent feature extraction and analysis, but also significantly reduces background noise and interference from irrelevant signals by suppressing signals in non-target frequency bands.

[0070] Time-frequency analysis technology, based on short-time Fourier transform, processes high-frequency vibration signals after bandpass filtering and calculates the following: ,in, It is a signal in the time domain. It revolves around the time window function. It is the angular frequency. This method allows us to obtain a detailed view of the frequency components of a signal as it changes over time.

[0071] Based on the time-frequency analysis results, high-frequency vibration characteristics related to the high-frequency vibration signal properties, including instantaneous energy, are extracted. Spectrum peak High-frequency energy distribution .

[0072] The instantaneous energy extraction calculation includes: within a given time window, integrating the squares of the amplitudes of all frequency components in the time-frequency representation obtained from the STFT, to calculate:

[0073] Instantaneous energy ; indicates that at a given time point t, the instantaneous energy of the signal is obtained by summing the squares of the amplitudes of all frequency components in the short-time Fourier transform (STFT) result at that time point, where, It is the result of short-time Fourier transform.

[0074] In bearing inspection, instantaneous energy can be used to identify sudden events or abnormal vibrations that occur during bearing operation, which usually indicate defects or damage in the bearing.

[0075] The peak frequency (PF) refers to the frequency corresponding to the point with the largest spectral amplitude within a specific time window. This value can represent the main frequency component of the vibration signal within that specific time window. The calculation method involves finding the point with the largest amplitude in each time window of the STFT result.

[0076] Spectrum peak ,in, Indicates time The frequency corresponding to the peak value of the spectrum. The operation is used to find out why Frequency of reaching the maximum value .

[0077] In bearing inspection systems, identifying spectral peaks helps determine the main vibration frequencies generated during bearing operation, which is crucial for analyzing the bearing's operating condition and identifying specific types of faults (ball or raceway damage).

[0078] High-frequency energy distribution describes the proportion of energy in the high-frequency region to the total energy during the entire signal processing process, reflecting the importance and distribution of high-frequency components in the vibration signal. It is obtained by calculating the sum of energy in the high-frequency region of the STFT result and comparing it with the sum of energy across the entire frequency range, as follows:

[0079] ;

[0080] ;

[0081] ,

[0082] in, It is the sum of energy in the high-frequency region of time. It is in time The sum of energy across the entire frequency range. It is the defined high-frequency region. and These are the minimum and maximum frequencies in STFT analysis. By comparing the ratio of energy in the high-frequency region to the total energy, the proportion and importance of high-frequency components in the signal are quantified. It refers to the high-frequency energy distribution, i.e., the high-frequency ratio.

[0083] High-frequency energy distribution analysis in bearing inspection helps identify minute changes caused by bearing defects such as cracks, corrosion, or wear. These changes are more easily detected at high frequencies, and changes in this ratio indicate changes in bearing condition or are precursors to failure.

[0084] The spatial positioning technology in the multi-source localization module is based on DOA estimation. It utilizes the direction of arrival estimation of the sensor array to perform multi-source localization, and uses the multi-signal classification MUSIC algorithm. The steps are as follows:

[0085] Calculate the covariance matrix of the received signal. ,in, Indicates time The signal vector received by the sensor array, It is the total number of time samples. Indicates conjugate transpose;

[0086] For covariance matrix Perform feature decomposition to obtain the signal subspace and the noise subspace;

[0087] Constructing the MUSIC spectral peak function: ,in, It is a direction vector, which depends on the direction of arrival of the wave. It is the eigenvector matrix of the noise subspace;

[0088] Scan all directions of arrival ,turn up The peak value, the location of which is the estimated direction of the vibration source;

[0089] By analyzing the DOA estimation results of each sensor for the same vibration source, the position coordinates of the vibration source are calculated using the triangulation method.

[0090] The construction of a graph neural network specifically includes:

[0091] Constructing a graph structure: Nodes in the graph structure represent the directions of the identified vibration sources, and the features of the nodes include the corresponding high-frequency vibration characteristics, i.e., instantaneous energy. Spectrum peak High-frequency energy distribution The weights of the edges are defined based on the spatial distance or similarity between the vibration sources;

[0092] Graph Neural Network (GNN) Model Design: Design a GNN model that takes a graph structure as input and learns the complex spatial relationships and characteristic information of vibration sources by aggregating and updating information on graph nodes.

[0093] Model training and validation: The GNN model is trained using the prepared graph structure dataset. Different fault types and their corresponding graph structures are used as inputs, and the fault source location is used as the label. The model parameters are optimized using loss functions and optimization algorithms, and the model performance is evaluated using the validation set.

[0094] Fault source identification and localization: After the model is trained, the vibration source direction and high-frequency vibration characteristics are processed through the same preprocessing and graph construction steps and then input into the trained GNN model. The model will predict the type and specific location of the fault based on the information of the entire graph structure, thereby realizing the comprehensive analysis of multiple vibration sources and the localization of the fault source.

[0095] The graph neural network is specifically represented as follows:

[0096] Graph structure definition: Constructing a graph ,in, It is a set of nodes, each node representing a vibration source, and the node characteristics include high-frequency vibration features. (Instantaneous energy, spectral peak value, and high-frequency energy distribution) and DOA estimation results, It is a set of edges, and the weights of the edges. Calculated based on the spatial distance or similarity between vibration sources;

[0097] Information propagation and aggregation: In each iteration, nodes From its neighboring nodes Represents a node The set of neighboring nodes collects information and updates its own feature representation, using the following calculation formula:

[0098] ,in, It is a node In the Feature representation of the next iteration, UPDATE and These are the update and aggregation functions, which are specifically defined according to different GNN variants, including GCN and GAT;

[0099] Readout layer: For each node (vibration source) in the graph, the readout function READOUT is used to aggregate information from the entire graph, resulting in a graph-level representation used for final fault type and location prediction. ,in, It is the final iteration number. It is a graph-level representation;

[0100] Fault prediction: using graph-level representation Predict the type and location of faults using one or more fully connected layers: ,in, It predicts the type and location of the fault. and These are the weights and biases of the fully connected layer, and softmax is used to normalize the output, making it represent a probability distribution.

[0101] In this way, GNN can combine spatial information and high-frequency vibration characteristics from multiple vibration sources to learn complex failure modes. Ultimately, the model can predict the probability of a specific failure type and infer the specific location of the failure based on the spatial location information of the vibration source, providing accurate guidance for bearing maintenance and fault repair.

[0102] The following is a specific example of using the above graph neural network (GNN) to predict the type and specific location of a fault:

[0103] In a bearing inspection, an array of four sensors is deployed to monitor the bearing's vibration. In one monitoring session, the system detects two significant vibration sources, which may be caused by different bearing faults, such as inner ring damage and outer ring damage. The goal is to use a Generative Neural Network (GNN) to predict the specific fault type and location of these two vibration sources.

[0104] Step 1: Data collection and preprocessing: High-frequency vibration feature extraction is performed on the signals captured by each sensor to obtain data on instantaneous energy, spectral peak value and high-frequency energy distribution. At the same time, DOA estimation is performed to determine the approximate direction of the two vibration sources.

[0105] Step 2: Construct the graph structure

[0106] Based on DOA estimation, we know the directions of vibration source A and vibration source B, and take these two vibration sources as two nodes in the figure. The node features include the corresponding high-frequency vibration features.

[0107] Considering the spatial layout of the sensor array, edges between nodes can be defined, and the weights of the edges can be based on the relative positions between vibration sources or the similarity of high-frequency vibration characteristics.

[0108] Step 3: GNN Model Application: Input the constructed graph into the pre-trained GNN model. Inside the model, the features of the nodes (vibration sources) are updated through information propagation and aggregation processes, while considering the spatial relationships between nodes; the readout function is used to aggregate the information of the entire graph to obtain a graph-level representation that integrates the information of all vibration sources and the relationships between them.

[0109] Step 4: Fault Prediction:

[0110] Using graph-level representation, the type and location of faults are predicted through fully connected layers and a softmax function. The model outputs the prediction results:

[0111] Vibration source A: Inner ring damage, located in the northeast direction of the bearing.

[0112] Vibration source B: outer ring damage, located in the southwest direction of the bearing.

[0113] In this example, by combining the high-frequency vibration characteristics of each vibration source with the DOA estimation results, the GNN can learn the complex relationships between vibration sources and the association between each vibration source and a specific fault type. This comprehensive analysis enables the model to not only predict the type of fault but also accurately pinpoint the specific location of the fault, thus providing powerful guidance for bearing maintenance. Through this method, even in complex situations where multiple vibration sources coexist, the system can effectively diagnose and locate faults.

[0114] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.

[0115] This invention is intended to cover all such substitutions, modifications, and variations falling within the broad scope of the claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A bearing quality production inspection system based on bearing manufacturing, characterized by, The method comprises the following modules: Rotary simulation and signal acquisition module: responsible for simulating the rotary motion of the bearing and collecting vibration signals under controlled conditions; Signal processing module: based on band-pass filtering, the captured vibration signals are pre-processed to enhance the high-frequency signal part while suppressing the signals of non-target frequency bands; Feature extraction and analysis module: using time-frequency analysis technology, the pre-processed high-frequency vibration signals are analyzed to reveal the frequency components of the high-frequency vibration signals over time. Based on the time-frequency analysis results, high-frequency vibration features related to the characteristics of the high-frequency vibration signals are extracted; Multi-source positioning module: based on spatial positioning technology, the direction of the vibration source is identified; Adaptive high-frequency vibration feature learning and intelligent source identification module: based on the direction of the vibration source identified by the spatial positioning technology and the high-frequency vibration features, a graph neural network is constructed, and after training, the fault source and fault type are located; The graph neural network construction specifically includes: Constructing graph structure: nodes in the graph structure represent identified vibration source directions, the features of the nodes include the corresponding high-frequency vibration features, including instantaneous energy , spectral peak , high-frequency energy distribution , the weight of the edge is defined according to the spatial distance or similarity between the vibration sources; Graph neural network (GNN) model design: a GNN model is designed to accept a graph structure as input, and through information aggregation and update on graph nodes, the complex spatial relationship between vibration sources and their respective feature information are learned; Model training and verification: using the prepared graph structure dataset, the GNN model is trained, different fault types and their corresponding graph structures are used as input, and the fault source position is used as label. Loss function and optimization algorithm are used to optimize model parameters, and validation set is used to evaluate the performance of the model; Fault source identification and positioning: after the model training is completed, the vibration source direction and high-frequency vibration features are pre-processed and graph constructed, and then input into the trained GNN model. Based on the information of the entire graph structure, the model predicts the type and specific location of the fault, realizing the comprehensive analysis of multiple vibration sources and the positioning of the fault source.

2. The bearing quality production detection system based on bearing manufacturing according to claim 1, characterized by, The rotary simulation and signal acquisition module specifically includes a rotary simulation submodule and a signal acquisition submodule; The rotary simulation submodule specifically includes: Rotary drive unit: the bearing under test is installed on a rotating shaft, and the rotating shaft is driven to rotate by a motor to simulate the rotary motion of the bearing under test in actual application; Load condition setting: simulating the axial or radial load that the bearing bears in actual work; Speed control unit: adjust the rotation speed to cover different working conditions from low speed to high speed; The signal acquisition submodule uses a sensor array to capture high-frequency vibration signals generated when the bearing is running.

3. The bearing quality production detection system based on bearing manufacturing according to claim 2, characterized by, The signal processing module specifically includes: A frequency range is predefined, which matches the frequency of the high-frequency vibration signals expected to be generated when the bearing is running. A band-pass filter is used to allow only signal components within the predefined frequency range to pass through, while suppressing other signal components below and above the frequency range.

4. The bearing quality production detection system based on bearing manufacturing according to claim 1, characterized by, The time-frequency analysis technology is based on short-time Fourier transform (STFT) to process the high-frequency vibration signals after band-pass filtering; Based on the time-frequency analysis result, high-frequency vibration characteristics related to the characteristics of the high-frequency vibration signal are extracted, including instantaneous energy , spectral peak , and high-frequency energy distribution .

5. The bearing quality production detection system based on bearing manufacturing according to claim 4, characterized by, The extraction calculation of instantaneous energy includes: in a given time window, the amplitude square of all frequency components in the time-frequency representation obtained by STFT is integrated to calculate: instantaneous energy ; represents the instantaneous energy of the signal at a given point in time by summing the squared magnitudes of all frequency components in the short-time Fourier transform (STFT) result at that point in time.

6. The bearing quality production detection system based on bearing manufacturing according to claim 4, wherein, The frequency spectrum peak calculation method is to find the maximum amplitude point in each time window of the STFT result: Spectrum peak ,in, Indicates time The frequency corresponding to the peak value of the spectrum. The operation is used to find out why Frequency of reaching the maximum value .

7. The bearing quality production detection system based on bearing manufacturing according to claim 4, wherein, The high-frequency energy distribution is obtained by calculating the sum of energy in the high-frequency region in the STFT result and comparing it with the sum of energy in the entire frequency range, and by comparing the proportion of energy in the high-frequency region and the total energy, the proportion and importance of the high-frequency component in the signal are quantified.

8. The bearing quality production detection system based on bearing manufacturing according to claim 4, wherein, The spatial positioning technology in the multi-source positioning module is based on DOA estimation, uses the direction of arrival estimation of the sensor array for multi-source positioning, uses the multiple signal classification (MUSIC) algorithm, and the steps are as follows: computing a covariance matrix of the received signals performing eigen decomposition on the covariance matrix to obtain a signal subspace and a noise subspace A MUSIC spectrum peak function is constructed, all directions of arrival are scanned, a peak value is found, and the peak value position is the estimated vibration source direction; Through the DOA estimation result, through the analysis of the DOA estimation results of each sensor on the same vibration source, the position coordinates of the vibration source are calculated by using the triangular positioning method.

9. The bearing quality production detection system based on bearing manufacturing according to claim 1, characterized by, The graph neural network is specifically represented as follows: Graph structure definition: construct graph wherein, is a set of nodes, each node represents a vibration source, and the node features include high-frequency vibration features and DOA estimation results, is a set of edges, and the weight of the edge is calculated based on the spatial distance or similarity between vibration sources; Information propagation and aggregation: At each iteration, nodes collect information from their neighbor nodes representations of nodes collect information from their neighbor nodes and update their own feature representations; The readout layer: for each node in the graph, the readout function READOUT is used to aggregate the information of the entire graph to obtain a graph-level representation for the final fault type and position prediction. Failure prediction: using graph-level representation Predicting the type and location of failure by one or more fully connected layers: where, is the predicted failure type and location, and are the weights and biases of the fully connected layers, and softmax is used to normalize the output so that it represents a probability distribution.

Images