Bearing fault diagnosis method and system based on signal segmentation network and fingerprint features
By using signal segmentation networks and fingerprint features, the accuracy problem of bearing fault diagnosis under varying speeds and strong background noise was solved, achieving high-precision fault diagnosis and visualization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-30
- Publication Date
- 2026-03-17
AI Technical Summary
Existing bearing fault diagnosis methods are ineffective in handling variable speeds, strong background noise, and early, subtle faults under low-speed and heavy-load conditions, especially in online diagnostic tasks, resulting in poor diagnostic performance.
A method based on signal segmentation network and fingerprint features is adopted. The acoustic emission signal is processed by constructing a signal segmentation network, including encoder and decoder, convolution module, pooling and upsampling layer. Combined with post-processing clustering algorithm, fault pulse features are extracted and fault fingerprint feature map is drawn for diagnosis.
High-precision bearing fault diagnosis was achieved under varying speed conditions, improving the adaptability and accuracy of the method and enhancing the visualization of faults.
Smart Images
Figure CN116793678B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of bearing fault diagnosis technology, and in particular to a bearing fault diagnosis method and system based on signal segmentation network and fingerprint features. Background Technology
[0002] Bearings, as one of the most frequently used components in mechanical systems, are among the most common causes of machine failure. Bearing failure often results in significant economic losses and even personal injury. Therefore, bearing health monitoring and fault diagnosis are of paramount importance. Although vibration-based bearing fault diagnosis has developed rapidly over the past 20 years and is currently a research focus, acoustic emission technology shows more satisfactory results than vibration-based methods for specific applications, such as low-speed and heavy-load conditions, and for specific diagnostic tasks, such as damage localization and quantitative assessment.
[0003] Acoustic emission (AE) technology has been widely used for structural health and condition monitoring of mechanical systems. The rapid release of transient elastic waves can be used to monitor various structural damages, such as crack propagation, plastic deformation, and corrosion, as well as faults in rotating machinery, such as fatigue failure, friction and wear, and rotor rubbing. The AE signals of faulty bearings exhibit two significant and unique characteristics: periodic characteristics based on a global signal scale, and burst event characteristics based on discrete energy. Global characteristics focus on the periodic pulse components associated with the fault, and subsequent analysis generally follows envelope analysis methods. Discrete energy characteristics typically require first extracting the pulse components from the waveform stream, which not only supports subsequent analysis and index calculations but also saves data storage capacity. In fact, the sampling frequency of AE signals is generally high, so it is necessary to consider the advantages of both characteristics simultaneously to adapt to online diagnostic tasks under varying operating conditions.
[0004] In practical applications, the goal is to make decisions using as few sensors as possible while ensuring real-time performance. Therefore, utilizing acoustic emission technology alone to perform variable speed diagnostics becomes a natural approach. Fingerprint feature recognition, used for acoustic emission flow analysis, aims to segment fault pulses from the raw signal and visualize their periodicity. It provides an inspiring framework for online bearing monitoring in practical engineering. However, the effectiveness of this online monitoring strategy is somewhat affected when dealing with challenges such as wider speed ranges, strong background noise, and early, subtle faults. Summary of the Invention
[0005] The purpose of this invention is to provide a bearing fault diagnosis method and system based on signal segmentation network and fingerprint features, which can achieve high-precision fault diagnosis under varying speed conditions.
[0006] A bearing fault diagnosis method based on signal segmentation network and fingerprint features, comprising:
[0007] S1. Obtain a historical acoustic emission dataset and construct a signal segmentation network based on the historical acoustic emission dataset; the signal segmentation network includes an encoder used as a shrinking path and a decoder used as a symmetric expanding path; the encoder includes several first convolutional modules and the decoder includes several second convolutional modules.
[0008] Both the first convolutional module and the second convolutional module include a first convolutional layer, a first normalized layer, a first activation layer, a second convolutional layer, a second normalized layer, and a second activation layer connected in sequence.
[0009] Each of the first convolutional modules is connected through a pooling layer, which is used to perform max pooling downsampling; each of the second convolutional modules is connected through an upsampling layer, which is used to perform upsampling convolution.
[0010] The encoder and the decoder are connected via a splicing module;
[0011] S2, The signal segmentation network is trained based on the historical acoustic emission dataset to obtain the trained signal segmentation network;
[0012] S3, acquire signals from the bearing under test to obtain acoustic emission signals; input the acoustic emission signals into the trained signal segmentation network to obtain segmented signals;
[0013] S4, the portion of the segmented signal that is greater than or equal to the screening threshold is taken as the impact portion, and the portion of the segmented signal that is less than the screening threshold is taken as the background portion;
[0014] S5, take the start time of the current impact portion and the start time of the previous impact portion in the segmented signal as the pulse time interval, take the pulse time interval as the vertical axis and the time as the horizontal axis, and draw a fault fingerprint feature map.
[0015] S6. Based on the fault fingerprint feature map, perform fault diagnosis on the bearing under test.
[0016] Preferably, the historical acoustic emission dataset is divided into a training set and a test set. The signal segmentation network is iteratively trained using the training set. After each iteration, the signal segmentation network is tested based on the test set, and its loss value is calculated based on the weighted intersection-over-conclusion (IoC) loss function. If the loss value is greater than or equal to a set loss value, iterative training continues. If the loss value is less than the set loss value, the trained signal segmentation network is obtained. The weighted IoC loss function is as follows:
[0017]
[0018] In the formula: WIoU represents the loss value, C represents the total number of categories. In bearing fault diagnosis, the signal is divided into noise and impact components, so it is set to 2, p ij p represents the probability of predicting class i as class j. ii p represents the probability of predicting class i as class i. ji ω represents the probability of predicting class j as class i. i Indicates the category-level weight. n i This represents the total number of pixels in the i-th class.
[0019] Preferably, the step of obtaining the screening threshold is as follows:
[0020] Calculate the signal strength of the segmented signal, and determine the filtering threshold based on the signal strength;
[0021] The formula for calculating signal strength is as follows:
[0022]
[0023] Where: ASL dB The signal strength is represented by N, the length of the segmented signal is N, x(n) represents the nth data point of the segmented signal, and 1μ represents the decibel reference unit.
[0024] Preferably, S6 specifically includes:
[0025] Calculate the Euclidean distance between any two points in the fault fingerprint feature map, and sort them in descending order to obtain the distance dataset;
[0026] The local density of each point is calculated based on the distance dataset, and points with a local density greater than a set density value are used as the initial cluster centers.
[0027] If the Euclidean distance between two initial cluster centers is greater than the distance setting value, then these two initial cluster centers are retained. If the Euclidean distance between two initial cluster centers is less than or equal to the distance setting value, then either initial cluster center is discarded and the other initial cluster center is retained, thus obtaining the final set of cluster centers.
[0028] Clustering is performed on each point in the fault fingerprint feature map based on the cluster center point set to obtain the clustering result;
[0029] Calculate the mean and variance of each cluster in the clustering results;
[0030] If the average values of each class are multiples of each other, and the variance of any class is less than the set variance value, then the bearing under test has failed; otherwise, the bearing under test has not failed.
[0031] This invention also provides a bearing fault diagnosis system based on signal segmentation networks and fingerprint features, comprising:
[0032] A data network module is used to acquire historical acoustic emission datasets and construct a signal segmentation network based on the historical acoustic emission datasets; the signal segmentation network includes an encoder used as a shrinking path and a decoder used as a symmetric expanding path; the encoder includes several first convolutional modules and the decoder includes several second convolutional modules.
[0033] Both the first convolutional module and the second convolutional module include a first convolutional layer, a first normalized layer, a first activation layer, a second convolutional layer, a second normalized layer, and a second activation layer connected in sequence.
[0034] Each of the first convolutional modules is connected through a pooling layer, which is used to perform max pooling downsampling; each of the second convolutional modules is connected through an upsampling layer, which is used to perform upsampling convolution.
[0035] The encoder and the decoder are connected via a splicing module;
[0036] The network training module is used to train the signal segmentation network based on the historical acoustic emission dataset to obtain the trained signal segmentation network.
[0037] The data acquisition and segmentation module is used to acquire signals from the bearing under test to obtain acoustic emission signals; the acoustic emission signals are then input into the trained signal segmentation network to obtain segmented signals.
[0038] The data filtering module is used to identify the portion of the segmented signal that is greater than or equal to a filtering threshold as the impact portion and the portion of the segmented signal that is less than the filtering threshold as the background portion.
[0039] The fingerprint feature module is used to take the start time of the current impact portion and the start time of the previous impact portion in the segmented signal as the pulse time interval, and use the pulse time interval as the vertical axis and time as the horizontal axis to draw a fault fingerprint feature map.
[0040] The fault diagnosis module is used to diagnose the faults of the bearing under test based on the fault fingerprint feature map.
[0041] Preferably, the historical acoustic emission dataset is divided into a training set and a test set. The signal segmentation network is iteratively trained using the training set. The network training module iteratively trains the signal segmentation network based on the training set. After each iteration, the network is tested using the test set, and the loss value is calculated based on the weighted intersection-over-conclusion (IoC) loss function. If the loss value is greater than or equal to a set loss value, iterative training continues. If the loss value is less than the set loss value, the trained signal segmentation network is obtained. The weighted IoC loss function is as follows:
[0042]
[0043] In the formula: WIoU represents the loss value, C represents the total number of categories. In bearing fault diagnosis, the signal is divided into noise and impact components, so it is set to 2, p ij p represents the probability of predicting class i as class j. ii p represents the probability of predicting class i as class i. ji ω represents the probability of predicting class j as class i. i Indicates the category-level weight. n i This represents the total number of pixels in the i-th class.
[0044] Preferably, the step of obtaining the screening threshold is as follows:
[0045] Calculate the signal strength of the segmented signal, and determine the filtering threshold based on the signal strength;
[0046] The formula for calculating signal strength is as follows:
[0047]
[0048] Where: ASL dB The signal strength is represented by N, the length of the segmented signal is N, x(n) represents the nth data point of the segmented signal, and 1μ represents the decibel reference unit.
[0049] Preferably, the fault diagnosis module specifically comprises:
[0050] Calculate the Euclidean distance between any two points in the fault fingerprint feature map, and sort them in descending order to obtain the distance dataset;
[0051] The local density of each point is calculated based on the distance dataset, and points with a local density greater than a set density value are used as the initial cluster centers.
[0052] If the Euclidean distance between two initial cluster centers is greater than the distance setting value, then these two initial cluster centers are retained. If the Euclidean distance between two initial cluster centers is less than or equal to the distance setting value, then either initial cluster center is discarded and the other initial cluster center is retained, thus obtaining the final set of cluster centers.
[0053] Clustering is performed on each point in the fault fingerprint feature map based on the cluster center point set to obtain the clustering result;
[0054] Calculate the mean and variance of each cluster in the clustering results;
[0055] If the average values of each class are multiples of each other, and the variance of any class is less than the set variance value, then the bearing under test has failed; otherwise, the bearing under test has not failed.
[0056] The effects of this invention are as follows:
[0057] The bearing fault diagnosis method based on signal segmentation network and fingerprint features in this invention takes into account the acoustic emission signal characteristics of actual bearing faults, including various factors such as electromagnetic noise caused by broadband response.
[0058] This invention proposes a bearing fault diagnosis framework based on semantic segmentation under varying speed conditions, which incorporates a signal segmentation network into a fingerprint feature diagnosis algorithm, effectively improving the adaptability and accuracy of the method.
[0059] This invention presents a bearing fault diagnosis method based on signal segmentation networks and fingerprint features, employing a post-processing clustering algorithm to improve the visualization of fingerprint feature results. Attached Figure Description
[0060] Figure 1 This is a flowchart of the bearing fault diagnosis method based on signal segmentation network and fingerprint features of the present invention;
[0061] Figure 2 This is a schematic diagram of the test bench structure in an embodiment of the present invention;
[0062] Figure 3 The acoustic emission signal collected in the embodiments of the present invention;
[0063] Figure 4 This is the time-frequency representation of the acoustic emission signal collected in this embodiment of the invention;
[0064] Figure 5 This is a schematic diagram of the signal segmentation network in an embodiment of the present invention;
[0065] Figure 6 This is a schematic diagram of the segmented signal obtained by applying a signal segmentation network in an embodiment of the present invention;
[0066] Figure 7 This is a fault fingerprint feature map in an embodiment of the present invention;
[0067] Figure 8 This is a schematic diagram of the fingerprint clustering results obtained by applying the multicenter density peak clustering algorithm in this invention;
[0068] Figure 9 This is a schematic diagram comparing the effects of the method of the present invention and the algorithm based on the threshold value in the embodiments. Detailed Implementation
[0069] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.
[0070] Figure 1 This is the flowchart of the bearing fault diagnosis method based on signal segmentation network and fingerprint features of the present invention. For example... Figure 1 As shown, this invention provides a bearing fault diagnosis method based on signal segmentation networks and fingerprint features, which includes:
[0071] Step S1: Obtain the historical acoustic emission dataset and construct a signal segmentation network based on it. After performing high-frequency sampling, truncation, and mean removal on the historical acoustic emission dataset, it is segmented into several training and test sets with a length of 6000.
[0072] like Figure 5 As shown, the signal segmentation network includes an encoder that acts as a shrinking path and a decoder that acts as a symmetric expanding path; the encoder includes several first convolutional modules, and the decoder includes several second convolutional modules. The encoder part records the global background and attributes of the time series data, while the decoder part recovers the individual category information for each data point.
[0073] Both the first and second convolutional modules include a first convolutional layer, a first normalized layer, a first activation layer, a second convolutional layer, a second normalized layer, and a second activation layer connected in sequence. In this embodiment, the parameters of the first and second convolutional layers are as follows: kernel size is 3×1, number of convolutions is 32, stride is 3, and padding size is 1.
[0074] To collect contextual information, each first convolutional module is connected through a pooling layer, which is used to perform max pooling downsampling; the pooling region size for max pooling is 2×1.
[0075] To restore the length of the output, the second convolutional modules are connected through an upsampling layer, which is used to perform upsampling convolution.
[0076] To provide sufficient resolution in semantic segmentation while enabling the signal segmentation network to capture features at both shallow and deep levels, the encoder and decoder are connected via a splicing module.
[0077] Step S2: Train the signal segmentation network based on the historical acoustic emission dataset to obtain a trained signal segmentation network. Specifically, iteratively train the signal segmentation network based on the training set. After each iteration, test the network using the test set and calculate the loss value using the weighted intersection-over-conclusion (IoC) loss function. If the loss value is greater than or equal to a set loss value, continue iterative training. If the loss value is less than the set loss value, the trained signal segmentation network is obtained. The weighted IoC loss function is as follows:
[0078]
[0079] In the formula: WIoU represents the loss value, C represents the total number of categories. In bearing fault diagnosis, the signal is divided into noise and impact components, so it is set to 2, p ij p represents the probability of predicting class i as class j. ii p represents the probability of predicting class i as class i. ji ω represents the probability of predicting class j as class i. i Indicates the category-level weight. n i This represents the total number of pixels in the i-th class.
[0080] Because the cross-union loss function in this invention is set with ω i By assigning greater weight to the impact component and correspondingly less weight to the background component, the background component can be effectively suppressed, thus highlighting the impact component.
[0081] Step S3: Acquire signals from the bearing under test to obtain acoustic emission signals; input the acoustic emission signals into a trained signal segmentation network to obtain segmented signals. The signal segmentation network suppresses the background portion of the acoustic emission signals to highlight the impact portion.
[0082] Step S4 involves designating the portion of the segmented signal that is greater than or equal to the filtering threshold as the impulse portion and the portion that is less than the filtering threshold as the background portion. In this embodiment, the filtering threshold is set to the signal strength plus 15 dB, and step S4 is performed once every 20,000 data points.
[0083] The formula for calculating signal strength is as follows:
[0084]
[0085] Where: ASL dBThe signal strength is represented by N, the length of the segmented signal is N, x(n) represents the nth data point of the segmented signal, and 1μ represents the decibel reference unit.
[0086] Step S5: Using the start time of the current impact portion and the start time of the previous impact portion in the segmented signal as the pulse time interval, plot the pulse time interval on the ordinate and time on the abscissa to create a fault fingerprint feature map. Since the calculated statistical pattern resembles a human fingerprint, it is called a fault fingerprint feature.
[0087] Step S6: Perform fault diagnosis on the bearing under test based on the fault fingerprint feature map.
[0088] Specifically, step S6 uses a multi-center density peak clustering algorithm to diagnose the fault in the bearing under test. Step S6 is as follows:
[0089] Calculate the Euclidean distance between any two points in the fault fingerprint feature map and sort them in descending order to obtain the distance dataset.
[0090] The local density of each point is calculated based on the distance dataset, and points with a local density greater than a set density value are used as initial cluster centers. The formula for calculating the local density is as follows:
[0091]
[0092] In the formula: ρ i Let y represent the local density of the i-th data point. i d represents the i-th point in the fault fingerprint feature map. c It is the cutoff distance, pct is a set parameter, and Q is the number of distance data in the distance dataset.
[0093] If the Euclidean distance between two initial cluster centers is greater than the set distance value, then both initial cluster centers are retained. If the Euclidean distance between two initial cluster centers is less than or equal to the set distance value, then either initial cluster center is discarded and the other initial cluster center is retained, thus obtaining the final set of cluster centers.
[0094] Clustering is performed on each point in the fault fingerprint feature map based on the cluster center point set to obtain the clustering result.
[0095] Calculate the mean and variance of each cluster in the clustering results.
[0096] If the average values of each class are multiples of each other, and the variance of any class is less than the set variance value, then the bearing under test has failed; otherwise, the bearing under test has not failed.
[0097] Example 1
[0098] Take a rolling bearing test bench as an example. Figure 2 As shown, the test bench includes a drive motor 1, a speed and torque measurement module 2, a coupling 3, a support bearing 4, a hydraulic loading system 5, a test bearing housing 6, a force sensor 7, a test bearing 8, and an acoustic emission sensor 9. The drive motor 1 has a rated voltage of 380V, a rated power of 3KW, a rated speed of 3000rpm, and a rated torque of 9.55N·m. The speed and torque measurement module 2 is used to measure the actual speed and torque. A set of couplings 3 and a set of support bearings 4 form the connection support module of the entire test platform. A radial load is applied to the vertically movable test bearing housing 6 through the hydraulic loading system 5, and the force sensor is used to measure the load magnitude. The test bearing 8 can be replaced with different bearings for testing. Acoustic emission signals are collected by the acoustic emission sensor 9 during the test. The specific parameters of the test bearing 8 are: bearing type is cylindrical roller bearing, bearing pitch diameter is 42.5mm, number of rollers is 15, roller diameter is 6.6mm, and contact angle is 0°. The bearing fault location is located on the outer ring, and the fault type is slight spalling of 1.5mm. In this embodiment, the speed of the drive motor 1 varies from 500 to 800 rpm, first accelerating from 500 rpm to 800 rpm, then decelerating back to 500 rpm, and maintaining a speed of 600 rpm and 700 rpm for a period of time. The acoustic emission sampling frequency is set to 200 kHz, and the preamplifier is selected as 40 dB.
[0099] The acquired acoustic emission signals are as follows Figure 3 As shown, its time-frequency expression is as follows: Figure 4 As shown, a large amount of electromagnetic interference generated by the drive motor was found, the amplitude of which overwhelmed any potential fault impacts.
[0100] The obtained segmented signal is as follows Figure 6 As shown, from Figure 6 As can be seen, the acoustic emission signal is divided into an impact component and a background noise component.
[0101] The obtained fault fingerprint feature map is as follows Figure 7 As shown, the clustering results are as follows: Figure 8 As shown.
[0102] The method of this invention is compared with a method that only uses threshold values to determine fingerprint features, and the clustering results are as follows: Figure 9 As shown, Figure 9 (a) represents the clustering result obtained by applying the method of the present invention. Figure 9 (c) represents the clustering results obtained by using only the threshold value to determine fingerprint features. Figure 9 (b) is a visualization diagram showing the clustering results obtained by applying the method of the present invention. Figure 9 (d) is a visual representation of the clustering results obtained by using only threshold values to determine fingerprint features. Figure 9 As can be seen, the method of the present invention can form relatively clear fault fingerprint features, while the method of judging fingerprint features by only using threshold value can observe some fingerprint features in the high-speed region, but its effect is not good in the low-speed region, and misjudgment of fault pulses will lead to blurring of the pattern.
[0103] This invention also provides a bearing fault diagnosis system based on signal segmentation networks and fingerprint features, comprising:
[0104] The data network module is used to acquire historical acoustic emission datasets and construct a signal segmentation network based on the historical acoustic emission datasets. The signal segmentation network includes an encoder that serves as a shrinking path and a decoder that serves as a symmetric expanding path. The encoder includes several first convolutional modules, and the decoder includes several second convolutional modules.
[0105] Both the first convolutional module and the second convolutional module include a first convolutional layer, a first normalized layer, a first activation layer, a second convolutional layer, a second normalized layer, and a second activation layer connected in sequence.
[0106] Each first convolutional module is connected through a pooling layer, which is used to perform max pooling downsampling; each second convolutional module is connected through an upsampling layer, which is used to perform upsampling convolution.
[0107] The encoder and decoder are connected via a splicing module.
[0108] The network training module is used to train the signal segmentation network based on historical acoustic emission datasets to obtain a trained signal segmentation network.
[0109] The data acquisition and segmentation module is used to acquire signals from the bearing under test to obtain acoustic emission signals; the acoustic emission signals are then input into a trained signal segmentation network to obtain segmented signals.
[0110] The data filtering module is used to identify the portion of the segmented signal that is greater than or equal to the filtering threshold as the impact portion and the portion of the segmented signal that is less than the filtering threshold as the background portion.
[0111] The fingerprint feature module is used to draw a fault fingerprint feature map by taking the start time of the current impact portion and the start time of the previous impact portion in the segmented signal as the pulse time interval, using the pulse time interval as the vertical axis and time as the horizontal axis.
[0112] The fault diagnosis module is used to diagnose faults in the bearing under test based on the fault fingerprint feature map.
[0113] Optionally, the historical acoustic emission dataset is divided into a training set and a test set. The signal segmentation network is iteratively trained using the training set. The network training module iteratively trains the signal segmentation network based on the training set. After each iteration, the network is tested using the test set, and the loss value is calculated based on the weighted intersection-over-conclusion (IoC) loss function. If the loss value is greater than or equal to a set loss value, iterative training continues. If the loss value is less than the set loss value, the trained signal segmentation network is obtained. The weighted IoC loss function is as follows:
[0114]
[0115] In the formula: WIoU represents the loss value, C represents the total number of categories. In bearing fault diagnosis, the signal is divided into noise and impact components, so it is set to 2, p ij p represents the probability of predicting class i as class j. ii p represents the probability of predicting class i as class i. ji ω represents the probability of predicting class j as class i. i Indicates the category-level weight. n i This represents the total number of pixels in the i-th class.
[0116] Optionally, the steps for obtaining the filtering threshold are as follows:
[0117] Calculate the signal strength of the segmented signal and determine the filtering threshold based on the signal strength;
[0118] The formula for calculating signal strength is as follows:
[0119]
[0120] Where: ASL dB The signal strength is represented by N, the length of the segmented signal is N, x(n) represents the nth data point of the segmented signal, and 1μ represents the decibel reference unit.
[0121] Optionally, the fault diagnosis module specifically includes:
[0122] Calculate the Euclidean distance between any two points in the fault fingerprint feature map and sort them in descending order to obtain the distance dataset.
[0123] The local density of each point is calculated based on the distance dataset, and points with a local density greater than the set density value are used as the initial cluster centers.
[0124] If the Euclidean distance between two initial cluster centers is greater than the set distance value, then both initial cluster centers are retained. If the Euclidean distance between two initial cluster centers is less than or equal to the set distance value, then either initial cluster center is discarded and the other initial cluster center is retained, thus obtaining the final set of cluster centers.
[0125] Clustering is performed on each point in the fault fingerprint feature map based on the cluster center point set to obtain the clustering result.
[0126] Calculate the mean and variance of each cluster in the clustering results.
[0127] If the average values of each class are multiples of each other, and the variance of any class is less than the set variance value, then the bearing under test has failed; otherwise, the bearing under test has not failed.
[0128] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.
Claims
1. A bearing fault diagnosis method based on signal segmentation network and fingerprint features, characterized in that, It comprises: S1, obtaining a historical acoustic emission data set, and constructing a signal segmentation network based on the historical acoustic emission data set; the signal segmentation network comprises an encoder serving as a contraction path and a decoder serving as a symmetric expansion path; the encoder comprises a plurality of first convolution modules, and the decoder comprises a plurality of second convolution modules; The first convolution module and the second convolution module each comprise a first convolution layer, a first normalization layer, a first activation layer, a second convolution layer, a second normalization layer and a second activation layer connected in sequence; Each of the first convolution modules is connected through a pooling layer for maximum pooling operation downsampling; each of the second convolution modules is connected through an upsampling layer for upsampling convolution; The encoder and the decoder are connected through a splicing module; S2, training the signal segmentation network based on the historical acoustic emission data set to obtain a trained signal segmentation network; S3, signal acquisition is performed on a bearing to be tested to obtain an acoustic emission signal; the acoustic emission signal is input into the trained signal segmentation network to obtain a segmented signal; S4, a part of the segmented signal greater than or equal to a screening threshold value is taken as an impact part, and a part of the segmented signal less than the screening threshold value is taken as a background part; S5, the starting time of the current impact part and the starting time of the previous impact part in the segmented signal are taken as a pulse time interval, the pulse time interval is taken as the ordinate, and the time is taken as the abscissa to draw a fault fingerprint feature map; S6, based on the fault fingerprint feature map, the bearing to be tested is diagnosed for faults.
2. The bearing fault diagnosis method based on signal segmentation network and fingerprint feature according to claim 1, characterized in that, The historical acoustic emission data set is divided into a training set and a test set, the training set is used for iterative training of the signal segmentation network, after each iteration of the signal segmentation network, the test set is used for testing, and a loss value is calculated based on a weighted intersection over union loss function, if the loss value is greater than or equal to a loss setting value, the iteration is continued, if the loss value is less than the loss setting value, the trained signal segmentation network is obtained; the weighted intersection over union loss function is as follows: In the formula: WloU represents a loss value, C represents the total number of categories, in the bearing fault diagnosis, the signal is divided into noise part and impact part, so it is set to 2, p ij represents the probability of predicting class i as class j, p ii represents the probability of predicting class i as class i, p ji represents the probability of predicting class j as class i, ω i represents the weight of the class level, n i represents the total number of pixels of the i-th class.
3. The bearing fault diagnosis method based on signal segmentation network and fingerprint feature according to claim 1, characterized in that, The screening threshold value is obtained as follows: The signal intensity of the segmented signal is calculated, and the screening threshold value is determined based on the signal intensity; The signal intensity calculation formula is as follows: where: ASL dB represents signal strength, N is the length of the segmented signal, x(n) represents the nth data point of the segmented signal, and 1 μ represents a decibel reference unit.
4. The bearing fault diagnosis method based on signal segmentation network and fingerprint feature according to claim 1, characterized in that, S6 is specifically: The Euclidean distance between any two points in the fault fingerprint feature map is calculated, and the points are arranged in descending order to obtain a distance data set; The local density of each point is calculated based on the distance data set, and the points with local density greater than a density setting value are taken as initial clustering center points; If the Euclidean distance between two initial clustering center points is greater than a distance setting value, the two initial clustering center points are retained, if the Euclidean distance between two initial clustering center points is less than or equal to the distance setting value, any one of the initial clustering center points is discarded and the other initial clustering center point is retained to obtain a final clustering center point set; Each point in the fault fingerprint feature map is clustered based on the clustering center point set to obtain a clustering result; Calculate the mean and variance of each class in the clustering result; If the mean of each class is in a multiple relationship and the variance of any class is less than a variance setting value, the bearing under test has a fault, otherwise the bearing under test has no fault.
5. A bearing fault diagnosis system based on signal segmentation network and fingerprint features, characterized in that, It comprises: The data network module is configured to obtain a historical acoustic emission data set and construct a signal segmentation network based on the historical acoustic emission data set; the signal segmentation network comprises an encoder serving as a contraction path and a decoder serving as a symmetric expansion path; the encoder comprises a plurality of first convolution modules, and the decoder comprises a plurality of second convolution modules; The first convolution module and the second convolution module each comprise a first convolution layer, a first normalization layer, a first activation layer, a second convolution layer, a second normalization layer and a second activation layer connected in sequence; The first convolution modules are connected through a pooling layer, and the pooling layer is configured to perform maximum pooling operation and down-sampling; the second convolution modules are connected through an up-sampling layer, and the up-sampling layer is configured to perform up-sampling convolution; The encoder and the decoder are connected through a splicing module; The network training module is configured to train the signal segmentation network based on the historical acoustic emission data set to obtain a trained signal segmentation network; The data acquisition and segmentation module is configured to acquire a signal of a bearing under test to obtain an acoustic emission signal; and input the acoustic emission signal into the trained signal segmentation network to obtain a segmented signal. The data screening module is configured to regard a part of the segmented signal greater than or equal to a screening threshold value as an impact part and regard a part of the segmented signal less than the screening threshold value as a background part. The fingerprint feature module is configured to regard a start time of a current impact part and a start time of a previous impact part in the segmented signal as a pulse time interval, regard the pulse time interval as a vertical coordinate and regard time as a horizontal coordinate to draw a fault fingerprint feature map. The fault diagnosis module is configured to perform fault diagnosis on the bearing under test based on the fault fingerprint feature map.
6. The bearing fault diagnosis system based on signal segmentation network and fingerprint feature according to claim 5, characterized in that, The historical acoustic emission data set is divided into a training set and a test set; the signal segmentation network is iteratively trained using the training set; the network training module iteratively trains the signal segmentation network based on the training set; after each iteration, the test set is tested, and a loss value is calculated based on a weighted intersection over union loss function; if the loss value is greater than or equal to a loss setting value, the iteration is continued; if the loss value is less than the loss setting value, the trained signal segmentation network is obtained; the weighted intersection over union loss function is as follows: In the formula: WloU represents the loss value, C represents the total number of categories, in the bearing fault diagnosis, the signal is divided into noise part and impact part, so it is set to 2, p ij represents the probability of predicting class i as class j, p ii represents the probability of predicting class i as class i, p ji represents the probability of predicting class j as class i, ω i represents the weight of the class level, n i represents the total number of pixels of the i-th class.
7. The bearing fault diagnosis system based on signal segmentation network and fingerprint feature according to claim 5, characterized in that, The screening threshold value is obtained as follows: Calculate the signal intensity of the segmented signal and determine the screening threshold value based on the signal intensity; The signal intensity calculation formula is as follows: where: ASL dB represents signal strength, N is the length of the segmented signal, x(n) represents the nth data point of the segmented signal, and 1 μ represents a decibel reference unit.
8. The bearing fault diagnosis system based on signal segmentation network and fingerprint feature according to claim 5, characterized in that, The fault diagnosis module is specifically configured as follows: Calculate the Euclidean distance between any two points in the fault fingerprint feature map and arrange the points in descending order to obtain a distance data set; Calculate the local density of each point based on the distance data set and regard a point with a local density greater than a density setting value as an initial clustering center point; If the Euclidean distance between two initial clustering center points is greater than a distance setting value, the two initial clustering center points are reserved, if the Euclidean distance between two initial clustering center points is less than or equal to the distance setting value, any one initial clustering center point is discarded and the other initial clustering center point is reserved, and a final clustering center point set is obtained; Each point in the fault fingerprint feature map is clustered based on the clustering center point set, and a clustering result is obtained; An average value and a variance of each class in the clustering result are calculated; If the average value of each class is in a multiple relationship and the variance of any class is less than a variance setting value, the bearing under test is in failure, otherwise the bearing under test is not in failure.