Sliding bearing fault detection method and system based on image processing
By establishing a shaft defect data set and training model, combined with the vibration amplitude data analysis in the operating state, the problem of insufficient accuracy of sliding bearing fault detection in the existing technology is solved, and efficient, accurate fault identification and real-time sliding bearings are achieved monitor.
Patent Information
- Application Number
- CN202411474755.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-22
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-10-22
AI Technical Summary
The existing sliding bearing fault detection method based on image processing is insufficient in the working state of the equipment, making it difficult to effectively identify the fault of sliding bearings, especially in places where they are invisible inside.
By establishing a shaft defect data set and training an axial defect detection model, the shaft defect of the sliding bearing connection shaft is initially detected; if there is no defect, it means normal; if there is a defect, further judgment is made. Combined with multiple image processing in the operating state, vibration amplitude data is obtained, vibration amplitude index is calculated, and sliding bearings are analyzed. This method includes model structures of multi-scale convolutional layer, SE module, feature pyramid layer, pooling layer, fully connected layer and output layer, and uses data enhancement strategies to improve model recognition capabilities.
Improve the accuracy and reliability of sliding bearing fault detection, and can clearly and accurately collect images in the running state of sliding bearings, real-time detection without disassembly or shutdown, reducing misjudgment, especially in normal wear and tear. maintenance and downtime operation.
Smart Images

Figure CN118982733A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a sliding bearing fault detection method and system based on image processing. Background Art
[0002] Sliding bearings are mechanical components that use sliding between friction surfaces to support rotating shafts, and are usually composed of a bearing seat and a journal. Compared with rolling bearings, sliding bearings do not have rolling elements, and their load-bearing and sliding friction performance rely on the lubricating oil film on the friction surface. Sliding bearings are often used in situations where smooth and silent operation is required, such as in equipment such as motors, fans and turbines. Their application scenarios cover many fields such as heavy industry, power generation equipment, automobiles and aerospace. They perform particularly well in environments such as low speed and high load, continuous long-term operation, and high temperature and high pressure. They have the advantages of strong load-bearing capacity, low friction and long life.
[0003] With the development of artificial intelligence and machine learning, more and more patents and literature have begun to use image processing-based methods for fault detection. The image processing-based fault detection method first uses an image acquisition device such as a camera to obtain the surface image of the sliding bearing, and then uses computer vision and image analysis algorithms to determine whether the sliding bearing has problems such as wear and cracks. The advantage of the image-based detection method is that it is non-contact and can detect the state of the sliding bearing in real time.
[0004] Most of the existing fault detection methods based on image processing collect images of sliding bearings in a static state and process and analyze them. The accuracy of fault detection for sliding bearings in a working state of the equipment needs to be further improved.
[0005] Therefore, a sliding bearing fault detection method and system based on image processing are proposed. Summary of the invention
[0006] The purpose of the present invention is to provide a sliding bearing fault detection method and system based on image processing. First, by establishing an axial surface defect data set and training an axial surface defect detection model, the axial surface defect detection of the sliding bearing connecting shaft in a non-operating state is performed. If there is no defect on the axial surface, it means that the sliding bearing is normal. If there is a defect on the axial surface, further judgment is performed; then, by acquiring multiple images of the sliding bearing in a running state and processing them, vibration amplitude data is obtained for analyzing the sliding bearing in a running state; then, in combination with other data such as operating time and rotation speed, a vibration amplitude index is obtained to analyze whether the sliding bearing in a running state has a fault.
[0007] To achieve the above object, the present invention provides the following technical solutions: A sliding bearing fault detection method based on image processing, comprising: Step S1: Acquire and process historical shaft surface data to obtain a shaft surface defect data set; train a shaft surface defect detection model according to the shaft surface defect data set to obtain a trained shaft surface defect detection model.
[0008] Furthermore, the axial surface defect data set includes: The historical axial surface data includes surface images of the sliding bearing connecting shaft collected at different angles, and the surface images include images of the sliding bearing connecting shaft at different periods during long-term operation; Shaft surface defects include: scratches, cracks, pits, wear and corrosion.
[0009] Furthermore, the shaft surface defect detection model includes: The structure of the shaft surface defect detection model includes an input layer, a multi-scale convolution layer, an SE module, a feature pyramid layer, a pooling layer, a fully connected layer, a Dropout layer and an output layer; the multi-scale convolution layer adopts , and The convolution kernel size; During the training phase, an enhancement strategy based on training loss adjustment is adopted, including: obtaining and processing the training loss to obtain a set of low recognition accuracy samples, performing data enhancement operations on the low recognition accuracy sample set to obtain enhanced low recognition accuracy samples, and training the shaft surface defect detection model based on the enhanced low recognition accuracy samples; the data enhancement includes adding noise, blurring and scaling.
[0010] Step S2: Acquire and process the axial surface images of the sliding bearing connecting shaft collected at different angles in a non-working state to obtain an axial surface image set.
[0011] Furthermore, the axial image set includes: On the front side of the sliding bearing, a coordinate system is established in the horizontal and vertical directions, and the axial surface image of the sliding bearing connecting shaft is collected from the 0°, 90°, 180° and 270° directions of the sliding bearing connecting shaft, and denoising, correction and image enhancement processing are performed to obtain the axial surface image set.
[0012] Step S3: the trained axial surface defect detection model acquires and processes the data in the axial surface image set to obtain an axial surface defect detection result.
[0013] Furthermore, the shaft surface defect detection result includes: The trained axial surface defect detection model obtains and processes the axial surface images in the axial surface image set to obtain a first axial surface defect detection result set; obtains and calculates the data in the first axial surface defect detection result set to obtain the axial surface defect detection result; Further, if the shaft surface defect detection result is that there is no defect, it means that the sliding bearing is operating normally and there is no fault; if the shaft surface defect detection result is that there is a defect, step S4 is performed.
[0014] Step S4: If the shaft surface defect detection result is that there is a defect, a working image of the sliding bearing in a running state is obtained and processed to obtain a working image set; the working image set is obtained and processed to obtain vibration amplitude data.
[0015] Furthermore, the working image includes: The image acquisition device is arranged on the front side of the sliding bearing to obtain the image of the sliding bearing under the running state. The working images of the sliding bearing are processed to obtain a working image set; the working image includes the sliding bearing and the sliding bearing connecting shaft.
[0016] Furthermore, the working image set includes: Acquire the working image and perform smoothing processing on the working image through a Gaussian filter to obtain a first working image; acquire the first working image and detect blur characteristics in the first working image through Fourier transform and process the image to obtain blur data; perform a deconvolution operation on the first working image according to the blur data to obtain a corrected working image; All working images are corrected to obtain the working image set.
[0017] Furthermore, the vibration amplitude data includes: Acquiring data from the working image set and processing it using a contour extraction algorithm to obtain a bearing fitting contour; Acquire the bearing fitting profile and process it to obtain the minimum circumscribed circle and the maximum inscribed circle of the bearing fitting profile; calculate the area between the minimum circumscribed circle and the maximum inscribed circle to obtain the offset area; Acquire the bearing fitting profile and process it according to the profile fitting algorithm to obtain a fitting circle; acquire the fitting circle and process it to obtain the position of the fitting circle center in the image; process all the data in the working image set to obtain a fitting circle center position set; The data of the fitting circle center position set are acquired and processed to obtain the average axis position, the maximum vibration amplitude, the standard deviation of the vibration amplitude and the mean value of the vibration amplitude.
[0018] Step S5: Acquire the vibration amplitude data and process it to obtain a vibration amplitude index. If the vibration amplitude index is less than or equal to a preset vibration amplitude index threshold, the sliding bearing is not faulty and is normal wear. If the vibration amplitude index is greater than the preset vibration amplitude index threshold, the sliding bearing is faulty.
[0019] Furthermore, the vibration amplitude index includes: The vibration amplitude data, rotation speed, operation time and operation temperature are obtained and processed to obtain the vibration amplitude index. The calculation formula of the vibration amplitude index is: ; in, represents the vibration amplitude index, Indicates the offset area The weight coefficient of Indicates the maximum vibration amplitude The weight coefficient of Indicates the mean vibration amplitude The weight coefficient of Indicates the standard deviation of vibration amplitude The weight coefficient of Indicates the speed The weight coefficient of Indicates the operating time The weight coefficient of Indicates the operating temperature The weight coefficient of .
[0020] A sliding bearing fault detection system based on image processing, comprising: A shaft surface defect detection model training module is used to obtain and process historical shaft surface data to obtain a shaft surface defect data set; train a shaft surface defect detection model according to the shaft surface defect data set to obtain a trained shaft surface defect detection model; An axial surface image acquisition module is used to acquire and process axial surface images of the sliding bearing connecting shaft acquired at different angles in a non-working state to obtain an axial surface image set; An axial surface defect detection module is used for the trained axial surface defect detection model to obtain and process the data in the axial surface image set to obtain an axial surface defect detection result; A vibration amplitude data acquisition module is used to obtain and process a working image of the sliding bearing in a working state to obtain a working image set if the shaft surface defect detection result is that a defect exists; obtain and process the working image set to obtain vibration amplitude data; The sliding bearing fault detection module is used to obtain and process the vibration amplitude data to obtain a vibration amplitude index. If the vibration amplitude index is less than or equal to a preset vibration amplitude index threshold, the sliding bearing is not faulty and is normal wear; if the vibration amplitude index is greater than the preset vibration amplitude index threshold, the sliding bearing is faulty.
[0021] Compared with the prior art, the present invention has the following beneficial effects: 1. When a sliding bearing fails, it will cause damage to the sliding bearing connecting shaft. By establishing a shaft surface defect data set and training a shaft surface defect detection model, a preliminary inspection of the sliding bearing connecting shaft is first performed. If there is no defect on the shaft surface, it means that the sliding bearing is normal. If there is a defect on the shaft surface, further judgment is performed. This indirect judgment method helps to identify whether there is a fault in the invisible part inside the sliding bearing. In the model training process, by identifying and data enhancing samples with low recognition accuracy, and then training the model, the model's ability to detect subtle defects can be improved.
[0022] 2. By detecting the blur characteristics in the image and then using the motion blur convolution kernel for deconvolution operations, motion blur can be effectively corrected and image quality can be improved. When the sliding bearing and the connecting shaft are in operation, high-quality images can still be captured clearly and accurately. By obtaining data such as the offset area and vibration amplitude of the shaft in operation, real-time detection of the sliding bearing is achieved without disassembling or shutting down the sliding bearing for inspection, and data support is provided for subsequent judgment of whether the sliding bearing is faulty.
[0023] 3. The calculation formula of the vibration amplitude index includes multiple key variables such as offset area, maximum vibration amplitude, vibration amplitude mean, standard deviation, rotation speed, operating time and operating temperature. By assigning appropriate weights to each variable, the health status of the sliding bearing in operation can be evaluated, and the actual working condition of the sliding bearing can be judged more accurately. By comparing the vibration amplitude index with the preset threshold, it is possible to simply and intuitively determine whether there is an abnormality in the sliding bearing, which helps to reduce misjudgment, especially when the equipment is under normal wear, avoid unnecessary maintenance and shutdown operations, and improve the accuracy of fault judgment. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 A flowchart of a sliding bearing fault detection method based on image processing provided in Example 1 of the present invention; Figure 2 A schematic diagram of the structure of the shaft surface defect detection model provided in Example 1 of the present invention; Figure 3 A structural schematic diagram of a sliding bearing fault detection system based on image processing provided in Example 2 of the present invention. DETAILED DESCRIPTION
[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0026] Example 1
[0027] A certain type of servo motor is equipped with a sliding bearing. In order to detect whether the sliding bearing on the servo motor after long-term use is faulty, a machinery company introduced a sliding bearing fault detection method based on image processing provided by the present invention. The method flow is as follows: Figure 1 shown.
[0028] Step S1: Acquire and process historical shaft surface data to obtain a shaft surface defect data set; train a shaft surface defect detection model according to the shaft surface defect data set to obtain a trained shaft surface defect detection model.
[0029] Furthermore, the axial surface defect data set includes: The historical shaft surface data includes surface images of the sliding bearing connecting shaft collected at different angles, specifically mainly images of the journal portion of the sliding bearing connecting shaft in a non-operating state; the surface images include image data of the sliding bearing connecting shaft under different working conditions and different time nodes during the long-term operation process, and these images reflect the cumulative impact of external factors such as friction and stress on the shaft surface on the surface state; Shaft surface defects include but are not limited to the following categories: scratches, cracks, pits, wear, corrosion and other abnormalities. Scratches are shallow surface lines or grooves on the shaft surface caused by friction with the sliding bearing or external force. Cracks are tiny cracks caused by stress concentration on the shaft surface, which may affect the structural strength of the sliding bearing. Pits are local depressions or grooves on the shaft surface caused by long-term use or external impact. Wear is the loss of the shaft surface caused by long-term friction with the sliding bearing. Corrosion is the chemical reaction between the shaft surface and the environment, which causes the shaft surface to be oxidized or have corrosion spots.
[0030] By collecting and storing historical axial surface image data at different times and angles, a comprehensive and rich axial surface defect dataset is constructed. The axial surface defect detection model is trained using a training dataset containing a variety of defect types, providing strong data support for model training.
[0031] Furthermore, the structure of the shaft surface defect detection model is as follows: Figure 2 As shown, it includes: input layer, multi-scale convolution layer, SE module, feature pyramid layer, pooling layer, fully connected layer, Dropout layer and output layer; During the training phase, an enhancement strategy based on training loss adjustment is adopted, including: obtaining and processing the training loss to obtain a set of low recognition accuracy samples, performing data enhancement operations on the low recognition accuracy sample set to obtain enhanced low recognition accuracy samples, and training the shaft surface defect detection model based on the enhanced low recognition accuracy samples; the data enhancement includes adding noise, blurring and scaling.
[0032] Furthermore, the input layer is responsible for receiving the input data, i.e., the axial images in the dataset. The multi-scale convolution layer has convolution kernels of various sizes, such as , and The convolution kernel of the input image is used to perform convolution operations and extract features of different scales in the axial image. Small-sized convolution kernels are used to capture local details, such as tiny scratches and cracks; large-sized convolution kernels are used to identify larger defects, such as wear, pits, and corrosion. Through multi-scale convolution, the model can capture defect features of different sizes more comprehensively. The SE module, also known as the Squeeze-and-Excitation module, weights the features of different channels, selectively emphasizes important features, and suppresses irrelevant or useless features, thereby enhancing the model's ability to detect subtle defects. The feature pyramid layer integrates information from different scales and levels to improve the model's detection effect on various defects. The pooling layer reduces the dimension of the feature map through maximum pooling or average pooling, reduces the amount of calculation, and retains the most important feature information. The fully connected layer maps the extracted features into high-dimensional vectors and further uses this information for classification. The Dropout layer randomly discards some neurons during the training process to prevent the model from overfitting and improve the generalization ability of the model. The output layer classifies defects according to the features extracted by the previous layers, and finally gives whether there are defects on the axial surface and the specific category of the defect.
[0033] Furthermore, during the training phase of the model, an adjustment and enhancement strategy based on training loss is introduced. After each training iteration, the loss value of each sample is calculated. The loss value reflects the prediction error of the model on the sample. A higher loss value indicates that the model does not accurately recognize the sample. According to the loss value, the training samples are divided into a set of samples with low recognition accuracy and a set of samples with high recognition accuracy. The set of samples with low recognition accuracy usually represents samples that are difficult for the model or contain complex features. Data enhancement operations are performed on the set of samples with low recognition accuracy to generate enhanced samples, which are then trained again to make the model perform better when dealing with defects that are difficult to identify.
[0034] Table 1 shows the training results of the shaft surface defect detection model. As the number of training rounds increases, the training loss and verification loss of the model gradually decrease, and the accuracy continues to improve.
[0035] Table 1. Training results
[0036] The model's axial surface defect detection accuracy is improved by enhancing and retraining samples with low recognition accuracy; the axial surface defect detection model realizes the detection and identification of axial surface defects through a series of operations such as multi-scale feature extraction, adaptive feature enhancement and multi-level information fusion. When dealing with complex axial surface defect detection tasks, it can more efficiently identify potential defects and improve overall detection accuracy and reliability.
[0037] Step S2: Acquire and process the axial surface images of the sliding bearing connecting shaft collected at different angles in a non-working state to obtain an axial surface image set.
[0038] Furthermore, the axial image set includes: On the front side of the sliding bearing, a coordinate system is established in the horizontal and vertical directions, and the axial surface image of the sliding bearing connecting shaft is collected from the 0°, 90°, 180° and 270° directions of the sliding bearing connecting shaft, and denoising, correction and image enhancement processing are performed to obtain the axial surface image set.
[0039] Furthermore, for the shaft surface in complex scenes, the sliding bearing connecting shaft can be cleaned first, and then the shaft surface images can be collected from more directions; after collecting the images, the collected images are processed by denoising, correction and image enhancement to obtain a high-quality shaft surface image set for shaft surface defect detection in subsequent steps.
[0040] Axial images are collected from multiple directions and angles to ensure full coverage of the entire axial surface and avoid missing certain areas due to limitations in the shooting angle. Images are processed to obtain high-quality images, ensuring data clarity and accuracy, and providing a solid foundation for subsequent axial surface defect detection.
[0041] Step S3: the trained shaft surface defect detection model acquires and processes the data in the shaft surface image set to obtain a shaft surface defect detection result; Furthermore, the shaft surface defect detection result includes: The trained axial surface defect detection model obtains and processes the axial surface images in the axial surface image set to obtain a first axial surface defect detection result set; obtains and calculates the data in the first axial surface defect detection result set to obtain the axial surface defect detection result; Further, the initial defect number is set to 0. In this example, there are 4 axial plane images in the axial plane image set. If the axial plane image has defects, the defect number is incremented by 1. For example, the data in the first axial plane defect detection result set shows that three axial plane images have defects and one axial plane image does not have defects, so the defect number is 3. Further, if the number of defects is greater than or equal to 2, it means that there are defects on the axial surface of the sliding bearing connecting shaft, and if the number of defects is less than 2, it means that there are no defects on the axial surface of the sliding bearing connecting shaft; Further, if the shaft surface defect detection result is that there is no defect, it means that the sliding bearing is operating normally and there is no fault; if the shaft surface defect detection result is that there is a defect, step S4 is performed.
[0042] Because internal faults of sliding bearings can cause defects in the axial surface of the connecting shaft, the axial surface integrity of the sliding bearing connecting shaft reflects from the side whether the sliding bearing is operating normally, and helps to identify whether there are faults in invisible places inside the sliding bearing. Detecting axial surface defects through images collected from multiple angles can effectively avoid misjudgment caused by occlusion or limitations of a single angle, thereby improving the accuracy and reliability of the detection results.
[0043] Step S4: If the shaft surface defect detection result is that there is a defect, a working image of the sliding bearing in a working state is obtained and processed to obtain a working image set; the working image set is obtained and processed to obtain vibration amplitude data.
[0044] Furthermore, the working image includes: The image acquisition device is arranged on the front side of the sliding bearing to obtain the image of the sliding bearing under working condition. The working images of the sliding bearing are processed to obtain a working image set; the working image includes the sliding bearing and the sliding bearing connecting shaft.
[0045] The image acquisition device is set on the front, which is conducive to obtaining a clear view of the sliding bearing and the connecting shaft. By combining multiple images from the front view, data support is provided for the subsequent comprehensive analysis of whether there is a fault in the sliding bearing under operation, which helps to reduce misjudgment caused by local occlusion or view deviation.
[0046] Furthermore, the working image set includes: The working image is acquired and smoothed by a Gaussian filter to obtain a first working image; the first working image is acquired and blur characteristics in the first working image are detected and processed by Fourier transform to obtain blur data; the blur data is acquired and deconvolution operation is performed on the first working image by a motion blur convolution kernel to obtain a corrected working image; the blur data includes a motion direction and a blur length.
[0047] All working images are corrected to obtain a working image set.
[0048] By smoothing the working image, the noise in the image can be effectively reduced, making the image clearer; by analyzing the blur characteristics through Fourier transform and using the motion blur convolution kernel for deconvolution processing, the image quality can be greatly improved, the influence of motion blur can be eliminated, and a clear and accurate working image can be obtained, which helps to improve the accuracy of subsequent bearing status analysis and fault detection.
[0049] Furthermore, the vibration amplitude data includes: Acquiring data from the working image set and processing it using a contour extraction algorithm to obtain a bearing fitting contour; Furthermore, the contour extraction algorithm can utilize existing technologies such as edge detection algorithms to process the working image, separate the surface contour of the bearing shell in the sliding bearing from the complex background, and obtain the bearing fitting contour.
[0050] Obtain the bearing fitting profile and process it to obtain the minimum circumscribed circle and the maximum inscribed circle of the bearing fitting profile; calculate the area between the minimum circumscribed circle and the maximum inscribed circle to obtain the offset area ; Furthermore, the bearing fitting profile is processed using existing technologies such as Hough transform and convex hull algorithm to obtain the minimum circumscribed circle and the maximum inscribed circle; the area between the minimum circumscribed circle and the maximum inscribed circle represents the deformation of the bearing bush of the sliding bearing in operation due to operation, and the offset is used as to indicate; to all The above processing is performed on the image to obtain a set of offsets. The offset area of the bearing bush of the sliding bearing in the operating state is calculated based on the data of the set of offsets. The calculation formula is: ; in, represents the offset area, Indicates the number of images, Indicates The offset calculated in the image.
[0051] Acquire the bearing fitting profile and process it according to the profile fitting algorithm to obtain a fitting circle; acquire the fitting circle and process it to obtain the position of the fitting circle center in the image; process all the data in the working image set to obtain a fitting circle center position set; Furthermore, after obtaining the bearing fitting profile, the profile fitting algorithm can process the bearing fitting profile using existing technology to obtain the fitting circle; further process the fitting circle to obtain the coordinate position of the center of the fitting circle in the image. , process all images and obtain the set of fitting circle center positions.
[0052] The data of the fitting circle center position set are acquired and processed to obtain the average axis position, the maximum vibration amplitude, the standard deviation of the vibration amplitude and the mean value of the vibration amplitude.
[0053] Furthermore, by calculating the mean of the coordinate positions in the set of fitted circle center positions, the average axis position can be obtained. , by calculating the error between the fitting center and the average position, the vibration amplitude can be obtained. The calculation formula is: ; in, Indicates The vibration amplitude in the image, Indicates The position coordinates of the center of the fitted circle in the image, Indicates the distance calculation method.
[0054] Furthermore, the above processing is performed on all images to obtain the vibration amplitude set , maximum vibration amplitude Represents the maximum value in the vibration amplitude set. The calculation formula for the mean vibration amplitude is: ; in, Represents the mean value of vibration amplitude; the calculation formula of vibration amplitude standard deviation is: ; in, Represents the standard deviation of vibration amplitude.
[0055] By extracting the fitting contour of the bearing and calculating the area between the minimum circumscribed circle and the maximum inscribed circle, the deformation degree of the bearing shell of the sliding bearing during operation is accurately quantified; the calculation of other vibration amplitude data such as the maximum vibration amplitude, the mean vibration amplitude and the standard deviation of the vibration amplitude provides the vibration amplitude change of the bearing under operation, realizing real-time monitoring of the health status of the bearing; using multiple images for data analysis reduces the impact of single image noise or error on the overall detection results, and improves the reliability of the vibration amplitude data.
[0056] Step S5: Acquire the vibration amplitude data and process it to obtain a vibration amplitude index. If the vibration amplitude index is less than or equal to a preset vibration amplitude index threshold, the sliding bearing is not faulty and is normal wear. If the vibration amplitude index is greater than the preset vibration amplitude index threshold, the sliding bearing is faulty.
[0057] Furthermore, the vibration amplitude index includes: The offset area, the vibration amplitude data, the rotation speed, the machine operation time and the operation temperature data are obtained and processed to obtain the vibration amplitude index. The calculation formula of the vibration amplitude index is: ; in, represents the vibration amplitude index, Indicates the offset area The weight coefficient of Indicates the maximum vibration amplitude The weight coefficient of Indicates the mean vibration amplitude The weight coefficient of Indicates the standard deviation of vibration amplitude The weight coefficient of Indicates the speed The weight coefficient of Indicates the operating time The weight coefficient of Indicates the operating temperature The weight coefficient of .
[0058] Furthermore, in this embodiment, the offset area is 2.403 mm 2 , the maximum vibration amplitude is 0.321mm, the mean vibration amplitude is 0.258mm, the standard deviation of the vibration amplitude is 0.103, the speed is 1500, the operating temperature is 86.24℃, the machine operation time is 302 hours, and finally after calculation, the vibration amplitude index of the sliding bearing is 188.06, which is greater than the preset vibration amplitude index threshold of 180, indicating that the sliding bearing is faulty.
[0059] The vibration amplitude index calculates a specific value by assigning different weight coefficients to key influencing factors such as offset area, vibration amplitude, temperature and operating time. It effectively quantifies the health status of the sliding bearing, avoids misjudgment caused by relying on only one factor, improves the overall decision-making efficiency, and facilitates quick judgment of whether the sliding bearing has abnormalities or faults.
[0060] Firstly, the shaft surface defect detection model is used to detect defects on the connecting shaft of the sliding bearing, which indirectly reflects whether the sliding bearing is operating normally and helps to identify whether there is a fault in the invisible part inside the sliding bearing. After the defect is detected, the system will enter the vibration data analysis stage of the sliding bearing. By acquiring multiple images of the sliding bearing in operation and analyzing and processing them, vibration amplitude data such as the offset area, maximum vibration amplitude and vibration amplitude standard deviation are obtained, providing a reliable basis for fault detection of the sliding bearing. Finally, the vibration amplitude data is combined with data such as the operating time to calculate the vibration amplitude index, so as to quickly and accurately determine whether the sliding bearing is faulty.
[0061] Example 2 like Figure 3 The structure diagram of a sliding bearing fault detection system based on image processing provided by the present invention is shown. First, the shaft surface defect detection model training module obtains and processes historical shaft surface data to obtain a shaft surface defect data set; then, the shaft surface defect detection model is trained according to the shaft surface defect data set to obtain a trained shaft surface defect detection model.
[0062] After the model training is completed, a sample set is randomly selected from the test set for testing. The results are shown in Table 2, which shows the recognition accuracy of various types of defects. The overall recognition accuracy is 95.33%.
[0063] Table 2. Defect recognition results
[0064] Through technologies such as multi-scale convolution, the shaft surface defect detection model can accurately identify different types of shaft surface defects; in particular, through targeted data enhancement and training, the model can maintain a high recognition accuracy of about 95.33% under complex working conditions, greatly improving the efficiency and accuracy of shaft surface defect detection.
[0065] The axial surface image acquisition module is used to acquire and process the axial surface images of the sliding bearing connecting shaft acquired at different angles in a non-working state to obtain an axial surface image set.
[0066] In the shaft surface defect detection module, the trained shaft surface defect detection model obtains the data in the shaft surface image set and analyzes and processes it to obtain the shaft surface defect detection result. If the shaft surface defect detection result shows that there is a defect, it enters the vibration amplitude data acquisition module.
[0067] If the shaft surface defect detection result is that there is a defect, the vibration amplitude data acquisition module obtains and processes the working image of the sliding bearing in the working state to obtain a working image set; obtains and processes the working image set to obtain vibration amplitude data.
[0068] Furthermore, the innovation of this embodiment lies in the introduction of new vibration amplitude data and the combination of more features to more accurately measure the operating state of the sliding bearing.
[0069] New vibration amplitude data include: base vibration frequency, dynamic offset, vibration directionality offset and temperature vibration coupling coefficient.
[0070] Firstly, the vibration spectrum of the sliding bearing in operation is obtained by Fourier transform, and the reference vibration frequency of the sliding bearing is extracted therefrom. The reference vibration frequency can reflect the vibration data of the sliding bearing under different speeds and loads.
[0071] By detecting the displacement difference of the axis center of the sliding bearing connecting shaft in each working image under the running state, the dynamic offset can be calculated. The calculation formula is: ; in, Indicates the dynamic offset, represents the number of working images, Indicates The axis position coordinates of the sliding bearing connecting shaft in the working image, Indicates the average axis center position of the axis center of the sliding bearing connection shaft in all working images.
[0072] The vibration directional deviation is used to quantify the deviation direction of the sliding bearing by detecting the main vibration direction of the sliding bearing during operation. The calculation formula is: ; in, Indicates the main vibration direction of the sliding bearing.
[0073] At high temperatures, bearing materials may deform slightly, causing different vibration responses. The temperature-vibration coupling coefficient combines the operating temperature and vibration data of the sliding bearing to evaluate the impact of temperature changes on bearing vibration, which can more comprehensively reflect the actual operating status of the bearing. The calculation formula for the temperature-vibration coupling coefficient is: ; in, represents the temperature vibration coupling coefficient, represents the number of working images, Indicates The distance between the axis center of the sliding bearing connecting shaft and the average axis center position in the working image, Indicates The temperature data of the sliding bearing in the running state of the working image, and Represents the corresponding weight coefficient.
[0074] The sliding bearing fault detection module acquires and processes the vibration amplitude data to obtain a vibration amplitude index. If the vibration amplitude index is less than or equal to a preset vibration amplitude index threshold, the sliding bearing is not faulty and is normal wear; if the vibration amplitude index is greater than the preset vibration amplitude index threshold, the sliding bearing is faulty.
[0075] Furthermore, by combining the above data, a new vibration amplitude index is obtained, and the calculation formula is: ; in, represents the new vibration amplitude index, Indicates the reference vibration frequency The weight coefficient of Indicates dynamic offset The weight coefficient of Indicates the main vibration direction of the sliding bearing The weight coefficient of Represents the temperature vibration coupling coefficient The weight coefficient of .
[0076] Table 3. Test results
[0077] Table 3 shows the fault detection results of a batch of sliding bearings. When the vibration amplitude index is greater than 100, it indicates that the corresponding sliding bearing is faulty. It can be seen that the method proposed in the present invention can quickly and conveniently perform fault detection on sliding bearings, thereby improving detection efficiency and accuracy.
[0078] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A sliding bearing fault detection method based on image processing, It is characterized in that include: Step S1: Obtain and process historical shaft surface data to obtain a shaft surface defect data set; Training an axial surface defect detection model according to the axial surface defect data set to obtain a trained axial surface defect detection model; Step S2: acquiring and processing axial surface images of the sliding bearing connecting shaft collected at different angles in a non-working state to obtain an axial surface image set; Step S3: the trained shaft surface defect detection model acquires and processes the data in the shaft surface image set to obtain a shaft surface defect detection result; Step S4: if the shaft surface defect detection result is that there is a defect, a working image of the sliding bearing in a working state is obtained and processed to obtain a working image set; the working image set is obtained and processed to obtain vibration amplitude data; Step S5: Acquire and process the vibration amplitude data to obtain a vibration amplitude index. If the vibration amplitude index is less than or equal to a preset vibration amplitude index threshold, the sliding bearing is not faulty; if the vibration amplitude index is greater than the preset vibration amplitude index threshold, the sliding bearing is faulty.
2. According to the sliding bearing fault detection method based on image processing according to claim 1, It is characterized in that The shaft surface defect data set includes: The historical axial surface data includes surface images of the sliding bearing connecting shaft collected at different angles, and the surface images include images of the sliding bearing connecting shaft at different periods during long-term operation; Shaft surface defects include: scratches, cracks, pits, wear and corrosion.
3. According to the sliding bearing fault detection method based on image processing according to claim 1, It is characterized in that The shaft surface defect detection model includes: The structure of the shaft surface defect detection model includes an input layer, a multi-scale convolution layer, an SE module, a feature pyramid layer, a pooling layer, a fully connected layer, a Dropout layer and an output layer; the multi-scale convolution layer adopts , and The convolution kernel size; During the training phase, an enhancement strategy based on training loss adjustment is adopted, including: obtaining and processing the training loss to obtain a set of low recognition accuracy samples, performing data enhancement operations on the low recognition accuracy sample set to obtain enhanced low recognition accuracy samples, and training the shaft surface defect detection model based on the enhanced low recognition accuracy samples; the data enhancement includes adding noise, blurring and scaling.
4. According to the sliding bearing fault detection method based on image processing according to claim 1, It is characterized in that The axial image set includes: The axial surface images of the sliding bearing connecting shaft are collected from different directions of the sliding bearing connecting shaft, and denoising, correction and image enhancement processing are performed to obtain the axial surface image set.
5. The sliding bearing fault detection method based on image processing according to claim 1, It is characterized in that The shaft surface defect detection results include: The trained axial surface defect detection model obtains and processes the axial surface images in the axial surface image set to obtain a first axial surface defect detection result set; obtains and processes the data in the first axial surface defect detection result set to obtain the axial surface defect detection result; If the shaft surface defect detection result is that there is no defect, it means that the sliding bearing is operating normally and there is no fault; if the shaft surface defect detection result is that there is a defect, step S4 is performed.
6. A sliding bearing fault detection method based on image processing according to claim 1, It is characterized in that The working image includes: The image acquisition device is arranged on the front side of the sliding bearing to obtain the image of the sliding bearing under the running state. The working images of the sliding bearing are processed to obtain a working image set; the working image includes the sliding bearing and the sliding bearing connecting shaft.
7. The sliding bearing fault detection method based on image processing according to claim 1, It is characterized in that The working image set includes: Acquire the working image and perform smoothing processing on the working image through a Gaussian filter to obtain a first working image; acquire the first working image and detect blur characteristics in the first working image through Fourier transform and process the image to obtain blur data; perform a deconvolution operation on the first working image according to the blur data to obtain a corrected working image; All working images are corrected to obtain the working image set.
8. The sliding bearing fault detection method based on image processing according to claim 1, It is characterized in that The vibration amplitude data includes: Acquiring data from the working image set and processing it using a contour extraction algorithm to obtain a bearing fitting contour; Acquire the bearing fitting profile and process it to obtain the minimum circumscribed circle and the maximum inscribed circle of the bearing fitting profile; calculate the area between the minimum circumscribed circle and the maximum inscribed circle to obtain the offset area; Acquire the bearing fitting profile and process it according to the profile fitting algorithm to obtain a fitting circle; acquire the fitting circle and process it to obtain the position of the fitting circle center in the image; process all the data in the working image set to obtain a fitting circle center position set; The data of the fitting circle center position set are acquired and processed to obtain the average axis position, the maximum vibration amplitude, the standard deviation of the vibration amplitude and the mean value of the vibration amplitude.
9. The sliding bearing fault detection method based on image processing according to claim 1, It is characterized in that The vibration amplitude index includes: The vibration amplitude data, rotation speed, operation time and operation temperature are obtained and processed to obtain the vibration amplitude index. The calculation formula of the vibration amplitude index is: ; in, represents the vibration amplitude index, Indicates the offset area The weight coefficient of Indicates the maximum vibration amplitude The weight coefficient of Indicates the mean vibration amplitude The weight coefficient of Indicates the standard deviation of vibration amplitude The weight coefficient of Indicates the speed The weight coefficient of Indicates the operating time The weight coefficient of Indicates the operating temperature The weight coefficient of .
10. A sliding bearing fault detection system based on image processing, It is characterized in that include: A shaft surface defect detection model training module is used to obtain and process historical shaft surface data to obtain a shaft surface defect data set; Training an axial surface defect detection model according to the axial surface defect data set to obtain a trained axial surface defect detection model; An axial surface image acquisition module is used to acquire and process axial surface images of the sliding bearing connecting shaft acquired at different angles in a non-working state to obtain an axial surface image set; An axial surface defect detection module is used for the trained axial surface defect detection model to obtain and process the data in the axial surface image set to obtain an axial surface defect detection result; A vibration amplitude data acquisition module is used to obtain and process the working image of the sliding bearing in a working state to obtain a working image set; obtain and process the working image set to obtain vibration amplitude data; The sliding bearing fault detection module is used to obtain and process the vibration amplitude data to obtain a vibration amplitude index. If the vibration amplitude index is less than or equal to a preset vibration amplitude index threshold, the sliding bearing is not faulty and is normal wear; if the vibration amplitude index is greater than the preset vibration amplitude index threshold, the sliding bearing is faulty.
Citation Information
Patent Citations
Method for real-time detecting bearing shell vibration by image analysis
CN101458119A
Fault detection method and related product
CN110602485A
Bearing multi-dimensional defect detection method and system
CN115144181A
Defect target area image positioning segmentation method for rolling contact fatigue defect
CN115239666A
Bearing fault diagnosis method and system
CN117312924A
Cited By
Image distortion correction method for bearing bush surface defect detection
CN120259146A
Intelligent detection system and method for train bogie
CN122130398A
