Vibration and image fused high pressure roller mill roller surface stud diagnostic system and method

Abstract

This invention relates to a diagnostic system and method for high-pressure roller mill roller surface pins based on vibration and image fusion. By fusing multimodal vibration and image data, a collaborative diagnostic system is constructed, and a joint decision-making mechanism with two modules is proposed. This enables vibration-image collaborative monitoring, rapid response to anomalies, and improved adaptive capabilities, thereby enhancing the safety and efficiency of industrial production. The vibration analysis module focuses on monitoring and warning of abnormal states from the perspective of equipment operation dynamic signals, while the image module focuses on identifying and locating physical defects from the perspective of roller surface appearance. The two modules complement each other, forming a complete fault diagnosis system, effectively overcoming the limitations of single-modal diagnostic methods in existing technologies. The diagnostic system and method of this invention improve the efficiency, accuracy, and real-time performance of roller surface pin fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment fault diagnosis, specifically to a vibration and image fusion diagnostic system and method for high-pressure roller mill roller surface pins. Background Technology

[0002] High-pressure roller mills are crucial equipment in industrial mineral processing. Their complex operating environment, large load fluctuations, and susceptibility to vibration, temperature changes, and variations in material properties all contribute to a high risk of malfunction. Common high-pressure roller mill failures include bearing wear, broken roller pins, and unstable operation. Failure to detect and address these issues promptly can lead to equipment downtime, severely impacting production efficiency and even posing safety hazards.

[0003] Traditional methods for diagnosing roller surface faults in high-pressure roller mills typically rely on manual inspection. However, manual inspection is limited by experience and subjective judgment, making it impossible to guarantee the comprehensiveness and accuracy of the inspection. Furthermore, it is not suitable for complex working conditions. In harsh environments such as high temperature, high pressure, and strong vibration, manual inspection is difficult to implement and carries high risks.

[0004] With the development of sensing technology, existing technologies have proposed methods for detecting roller surface pin faults, such as vibration analysis and image recognition. These methods can improve the automation and accuracy of fault detection to a certain extent, but they still have the following shortcomings.

[0005] Traditional vibration analysis techniques have weak anti-interference capabilities: Under high dust and strong vibration conditions, vibration signals are easily affected by material impact, transmission system noise, and foundation resonance, leading to confusion in the characteristic spectrum. For example, the weak impact signal generated by the breakage of the column nail is often masked by the rupture of air bubbles in the return material or the fluctuation of the roller gap, resulting in a high false alarm rate. Furthermore, it is not sensitive enough to low-amplitude faults such as early micro-cracks, resulting in a significant missed detection rate.

[0006] Single-modal image recognition is severely constrained by the environment: high reflectivity of the roller surface, dust obscuring, strong light reflection and high-speed movement cause image blurring and reduced contrast. Traditional image algorithms have difficulty in reliably extracting subtle features such as broken pins and peeling of the roller surface. In environments with high dust concentration, the effective recognition distance of visual sensors is shortened, which cannot meet the needs of online continuous monitoring.

[0007] In addition, sensors have short lifespans and fail quickly in harsh environments: high dust environments can easily cause optical lenses to wear out, and contact sensors are prone to rapid failure due to metal particle erosion, which can create a vicious cycle of "sensor failure → inaccurate diagnosis → maintenance delay → equipment damage". Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a vibration and image fusion diagnostic system and method for high-pressure roller mill surface pins. This system overcomes the problem of inaccurate fault feature diagnosis in existing high-pressure roller mill surface pin failure diagnosis. By fusing vibration and image detection, a collaborative diagnostic system and method are constructed. By introducing the SVDD anomaly detection model and deep neural network, accurate identification of early minute changes and complex structural problems is achieved, effectively overcoming the limitations of single-modal diagnostic methods in existing technologies. Furthermore, this invention proposes image quality assessment and lens cleaning strategies to extend sensor lifespan and improve the accuracy of collected data.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A vibration and image fusion diagnostic system for high-pressure roller mill roller surface pins includes a data acquisition system, a data processing system, a fault diagnosis system, and an alarm control system. The data acquisition system comprises an image acquisition module and a vibration acquisition module. The image acquisition module acquires images of the roller surface using an industrial camera installed near the roller surface, while the vibration acquisition module acquires vibration signals using acceleration vibration sensors located near the bearing seats at both ends of the roller. The data processing system performs multi-dimensional preprocessing and feature extraction on the image data and vibration signals. The fault diagnosis system includes a vibration analysis module and an image recognition module. The vibration analysis module constructs a vibration anomaly detection model based on Support Vector Data Description (SVDD) based on the extracted vibration signal features to identify structural anomalies caused by early, minor changes. The image recognition module employs an image anomaly detection model based on a deep learning network to achieve accurate identification and spatial localization of structural problems on the roller surface. When the fault diagnosis system detects a fault in the equipment, it transmits the relevant fault data to the alarm control system. The alarm control system sends an alarm signal to the host computer through the control module and triggers the image acquisition system to perform re-acquisition and image reconstruction to confirm the fault status.

[0010] Furthermore, the diagnostic system also includes a data quality monitoring system. The data quality monitoring system monitors the quality indicators of each frame of image in real time through a built-in image sharpness evaluation algorithm and dynamically compares them with the working threshold of the lens transmittance factor set by the system. If the image is found to be blurry or the transmittance drops beyond the preset working threshold, it can automatically determine that there is dust, dirt or other obstructions on the lens surface and trigger a cleaning mechanism. After cleaning, the system automatically performs a data quality re-check and confirms whether the cleaning effect meets the standard before deciding whether to re-acquire image data.

[0011] A method for diagnosing roller surface pin faults in high-pressure roller mills using vibration and image fusion, comprising the following steps: S1. System parameter initialization: The vibration signals and roller surface image data at both ends of the roller are collected through the data acquisition system. S2. The vibration signal is denoised by applying a time-domain synchronous averaging algorithm to the data processing system, and frequency domain features are extracted by a frequency domain analysis algorithm. Then, the signals are input into the pre-trained vibration anomaly detection model based on support vector data description (SVDD) in the vibration analysis module for detection. The detection results of four measuring points at both ends of the moving roll and the fixed roll are obtained. Then, the DS evidence fusion theory is used to fuse the detection results of the four measuring points to obtain the vibration anomaly detection report of the moving roll and the fixed roll. S3. If the vibration analysis module detects an abnormality on the roller surface, the image recognition module is used for further verification. After the data processing system preprocesses the collected roller surface image, it is input into the image anomaly detection model based on a deep learning network, and the fault type and spatial location result of the roller surface pin are output. S4. If the fault diagnosis system detects a fault in the equipment, it will trigger the alarm control system to send an audible and visual alarm to the host computer and link the data acquisition system to re-acquire and verify the data, thereby confirming the fault status.

[0012] Furthermore, in step S2, the frequency domain features extracted by the frequency domain analysis algorithm include the margin factor, kurtosis, and root mean square frequency.

[0013] Furthermore, in step S2, the DS evidence fusion theory calculates the conflict factor between the diagnostic results at both ends and excludes the completely conflicting diagnostic results as "interference between the gearbox and the motor", thereby obtaining a more accurate diagnostic result.

[0014] Furthermore, in step S2, the vibration signal is denoised using a time-domain synchronous averaging algorithm, with the specific formula as follows: ; In the formula, y(n) is the original signal. Let N be the denoised signal, N be the number of averages, k be the summation index, L be the length of the truncated window, and T be the window sliding step size. Both L and T are integer multiples of the rotation period.

[0015] Further, in step S3, the data processing system characterizes the image stability change caused by abnormal roller surface vibration during image acquisition using a jitter index. The jitter index is determined by extracting displacement differences from consecutive image frames. The formula for the jitter index is: ; In the formula, The average displacement, i.e., the degree of chattering. To monitor the number of frames within a period, For the first The displacement difference between the first frame and the next frame.

[0016] Furthermore, the fault diagnosis system calculates the image jitter index to achieve image quality assessment and adaptive filtering. When assessing image quality, a threshold is set. When the jitter index exceeds the preset threshold, it indicates that there is significant jitter in the current image sequence, and the image is deemed unusable and re-acquired. When the jitter index is below the threshold, it indicates that the image acquisition process is stable and the quality is qualified.

[0017] Furthermore, during the image data acquisition process, the image sharpness evaluation algorithm built into the data quality monitoring system monitors the quality index of each acquired image frame in real time and dynamically compares it with the working threshold of the lens transmittance factor set by the system. If the image quality index is lower than the preset working threshold, it can be automatically determined that there is dirt on the surface of the industrial camera lens and trigger the cleaning mechanism to clean the lens. After cleaning is completed, the image quality index is recalculated. If the cleaning effect is confirmed to meet the standard, image acquisition is resumed.

[0018] Furthermore, the working threshold for the lens transmittance factor is established according to the following formula: In the formula, The working threshold, This is the time interval since the last dust removal. The preset dust removal time interval, For time scale parameters, Transmittance measured in real time. Transmittance is for reference or ideal conditions. To detect comprehensive dust indicators from multiple angles, The minimum amount of dust to begin to attract attention. The scaling parameters for the air soot blowing response curve. , , These are the weighting coefficients for the timed dust removal factor, the light transmittance factor, and the multi-angle air blowing factor, respectively, and the sum of their coefficients is 1.

[0019] Beneficial effects This invention constructs a collaborative diagnostic system by integrating vibration and image multimodal data, and proposes a joint decision-making mechanism with two modules, effectively overcoming the limitations of single-modal diagnostic methods in the prior art. The diagnostic system and method of this invention can achieve vibration-image collaborative monitoring, rapid response to anomalies, and improved adaptive capabilities, thereby enhancing the safety and efficiency of industrial production. Among them, the vibration analysis module focuses on monitoring and early warning of abnormal states from the perspective of equipment operation dynamic signals, while the image module focuses on identifying and locating physical defects from the perspective of roller surface appearance. The two complement each other and together constitute a complete fault diagnosis system, improving the efficiency, accuracy, and real-time performance of roller surface pin fault diagnosis.

[0020] This invention introduces an SVDD vibration anomaly detection model and a deep neural network image anomaly detection model into the fault diagnosis module. The SVDD vibration anomaly detection model can learn and classify the change patterns between samples, and is suitable for identifying structural anomalies caused by early, minute changes. It is efficient, fast, and has strong immediacy. The deep neural network image anomaly detection model can achieve accurate identification and spatial positioning of structural problems such as broken, missing, and worn roller pins through end-to-end training. The combination of the two can achieve accurate identification and positioning of early, minute changes and complex structural problems, overcoming the shortcomings of traditional threshold analysis in lacking adaptability and multi-fault coupling diagnosis capabilities, and improving the accuracy and reliability of diagnosis.

[0021] This invention employs a high-resolution industrial camera and an acceleration vibration sensor for combined data acquisition, solving the problems of interference in traditional vibration analysis and environmental constraints on image recognition. To address sensor failure caused by harsh environments, protective structures such as dust covers and nozzles are introduced, and a data quality monitoring system is designed to monitor the quality of acquired images in real time. When unqualified image quality is detected, a cleaning mechanism is triggered in a timely manner to clean the lens, effectively improving image quality and extending the sensor's lifespan. Attached Figure Description

[0022] Figure 1 This is a structural block diagram of the high-pressure roller mill roller surface pin diagnosis system in this embodiment.

[0023] Figure 2 This is a schematic diagram of the installation structure of the high-pressure roller mill roller surface pin diagnostic system in this embodiment, wherein, Figure 2 (c) is a schematic diagram of the roller structure of a high-pressure roller mill. Figure 2 (a) is a schematic diagram of the industrial camera mounting structure. Figure 2 (b) is Figure 2 (a) is the bottom view.

[0024] Figure 3 This is a flowchart illustrating the operation of the high-pressure roller mill roller surface pin diagnosis method in this embodiment.

[0025] Figure 4 This is a schematic diagram of the image recognition module in this embodiment.

[0026] Figure 5 This is a schematic diagram of the vibration analysis module in this embodiment.

[0027] Reference numerals: 1-Drive motor, 2-Reducer, 3-Roller pin, 4-Roller, 5-Nozzle bracket, 6-Nozzle, 7-Camera lens shroud, 8-Camera, 9-Dust cover, 10-Camera bracket. Detailed Implementation

[0028] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0029] like Figure 1 As shown, this embodiment provides a high-pressure roller mill roller surface pin diagnosis system that integrates vibration and image, including: a data acquisition system, a data processing system, a fault diagnosis system, an alarm control system, and a data quality monitoring system.

[0030] (1) Data acquisition system: This system is mainly responsible for synchronously acquiring images and vibration signals during the operation of the high-pressure roller mill. It includes an image acquisition module and a vibration acquisition module. The image acquisition module uses a high-resolution industrial camera installed near the roller surface, along with an auxiliary light source and a slide rail bracket. It can flexibly adjust the shooting angle and distance to acquire images of the roller surface during the operation of the high-pressure roller mill, ensuring that clear and high-quality roller surface images are acquired and stored in lossless compressed PNG format for subsequent analysis. The vibration acquisition module uses high-sensitivity acceleration vibration sensors arranged near the bearing seats at both ends of the rollers (including the moving roller and the fixed roller). These acceleration vibration sensors have a high sampling rate and high response capability, and can accurately sense minute vibration signals during operation. They transmit the raw data to the data processing system in real time and save it in CSV format for subsequent spectrum analysis and pattern recognition. The data acquisition system supports parallel acquisition and time alignment of image and vibration data, laying the foundation for multimodal diagnostics.

[0031] (2) Data processing system, which undertakes the multi-dimensional preprocessing and feature extraction of image data and vibration raw data, and transmits the processed data to the fault diagnosis system for data analysis.

[0032] Image data is first preprocessed and converted into a data format suitable for deep learning models. High-quality datasets are generated using image enhancement techniques (such as rotation, flipping, and cropping) for model training. The data processing system also divides the processed data into training and testing sets to ensure that the model can be effectively evaluated during both training and testing phases.

[0033] The vibration data is initially denoised using time-domain synchronous averaging technology. Then, key statistical indicators such as margin factor, kurtosis, and root mean square frequency are extracted using frequency domain analysis algorithms such as Fast Fourier Transform (FFT). These features are used by the fault diagnosis system to build an anomaly detection model based on SVDD (Support Vector Data Description), thereby achieving early fault identification of vibration signals.

[0034] Among them, the fault diagnosis based on vibration data uses a time-domain synchronous averaging algorithm to denoise the sample data, and the formula used is: ; In the formula, y(n) is the original signal. Let N be the denoised signal, N be the number of averages, k be the summation index, L be the length of the truncated window, and T be the window sliding step size. Both L and T are integer multiples of the rotation period.

[0035] In addition, the breakage of the pins will cause unevenness on the roller surface, and the roller surface will generate irregular impacts and vibrations during operation. These abnormal vibrations will lead to an increase in the jitter index. The data processing system determines the jitter index by extracting displacement differences from continuous image frames.

[0036] The vibration index is determined according to the following formula: In the formula, This represents the average displacement (degree of chattering). To monitor the number of frames within a period, For the first The displacement difference between the first frame and the next frame.

[0037] The jitter index calculated in this invention has two functions. Firstly, the jitter index characterizes changes in image stability caused by abnormal roller surface vibration during image acquisition, obtained by analyzing displacement differences between consecutive image frames. When the jitter index exceeds a preset threshold, it indicates significant jitter in the current image sequence, affecting image clarity and geometric consistency, making it difficult to accurately reflect the true state of the roller surface. When the jitter index is below the threshold, it indicates stable image acquisition, and the acquired images meet the quality requirements for subsequent image processing and dataset construction. Therefore, this invention uses the jitter index to determine the quality of acquired images and selects data samples suitable for model training and state analysis. Secondly, the jitter index can assist in judging equipment vibration. After a roller surface nail breaks, the roller surface becomes uneven, and the vibration during equipment operation will be greater. The jitter index can determine the equipment vibration, capturing irregular impact vibrations caused by broken nails, uneven roller surfaces, etc., during equipment operation, issuing early warnings in the early stages of a fault to prevent the fault from worsening, thus playing a certain auxiliary role in the final diagnostic results.

[0038] (3) Fault diagnosis system. This system includes a vibration analysis module and an image recognition module, which are used to perform fault diagnosis analysis on the processed data transmitted by the data processing system. If a fault is diagnosed in the equipment, the fault result is transmitted to the alarm control system to perform fault-related operations. This system integrates vibration analysis and image recognition algorithms to build a high-precision and high-robust composite fault diagnosis capability.

[0039] The vibration analysis module constructs a vibration anomaly detection model based on SVDD (hereinafter referred to as the SVDD model) based on the feature parameters (margin factor, kurtosis, root mean square frequency, etc.) extracted from the vibration signal by the aforementioned data processing system. It can learn and classify the change patterns between samples and is suitable for identifying structural anomalies caused by early and minute changes.

[0040] like Figure 5 As shown, the vibration analysis module constructs a dataset by sampling data from a healthy high-pressure roller mill, and performs offline training on the SVDD model. Offline training is used to determine the distribution of vibration response data sample points of the target roller mill in a healthy state, thus providing a basis for the establishment of the SVDD model. The established SVDD model is a hypersphere that can describe the distribution range of healthy samples in the sample space, and can be used for online testing of roller mill vibration signals. During the online testing process, the vibration signals of the high-pressure roller mill under test are collected through the data acquisition system. The same data processing and sample segmentation methods as in offline training are used to obtain the test samples, and the SVDD model is used to evaluate whether these test samples are within the healthy range. Then, the data from four measurement points at both ends of the moving roller and the stationary roller are fused using DS evidence to output the detection results.

[0041] In practical applications, considering the differences between different roller presses (i.e., high-pressure roller mills), it is best for healthy samples and test samples to come from the same roller press. Therefore, the data from the roller press under test within a certain period of time after maintenance can be used for the offline training part.

[0042] DS evidence fusion is mainly used to address the problem that the roller drive end is easily affected by interference from the gearbox and motor. By calculating the conflict factor of the diagnostic results at both ends, the completely conflicting diagnostic results are excluded as "interference from the gearbox and motor", thus obtaining a more accurate diagnostic result.

[0043] like Figure 4As shown, the image recognition module adopts a deep neural network architecture (such as YOLO, ResNet, etc.) to achieve accurate identification and spatial positioning of structural problems such as broken, missing, and worn roller pins through end-to-end training. The general process is as follows: the image acquisition module collects image data during the operation of the high-pressure roller mill, and after storage, transmission and preprocessing, it calls the pre-trained deep learning model and inputs the image data into the model to realize the identification and positioning of the fault.

[0044] In actual fault detection, the vibration analysis module and the image recognition module work together. When the vibration analysis module identifies an abnormality on the roller surface, the image recognition module is used for further verification to avoid problems such as misjudgment and missed judgment that may occur with single-mode recognition.

[0045] (4) Data quality monitoring system. This system is mainly used to ensure the clarity and stability of image data. It monitors the quality indicators of each frame of image in real time through the built-in image clarity evaluation algorithm and compares them with the lens transmittance factor set by the system. If the image is found to be blurry or the transmittance drops beyond the preset working threshold, it can automatically determine that there is dust, dirt or other obstructions on the lens surface and trigger the cleaning mechanism. After cleaning, the system automatically performs data quality re-inspection and confirms whether the cleaning effect meets the standard before deciding whether to re-collect image data.

[0046] The operating threshold is established according to the following formula: In the formula, The operating threshold for the data quality monitoring system. This is the time interval since the last dust removal. The preset dust removal time interval, For time scale parameters, Transmittance measured in real time. Transmittance is for reference or ideal conditions. To detect comprehensive dust indicators from multiple angles, The minimum amount of dust to begin to attract attention. The scaling parameters for the air soot blowing response curve. , , These are the weighting coefficients for the timed dust removal factor, the light transmittance factor, and the multi-angle air blowing factor, respectively, with the sum of each weighting coefficient being 1.

[0047] like Figure 2The diagram shows the installation position of the industrial camera and the high-pressure roller mill of the present invention. The high-pressure roller mill includes two rollers 4 (moving roller and fixed roller). The drive motor 1 is connected to the rollers 4 through the reducer 2. The rollers 4 have pins 3 distributed on their surface. The industrial camera 8 is mounted on the rollers 4 through the camera bracket 10. The camera lens is provided with a lens cover 7. The industrial camera 10 is provided with a dust cover 9. The nozzle bracket 5 and nozzle 6 are provided below the camera lens, which can clean the lens, dust cover, etc.

[0048] The data quality monitoring system works in conjunction with the data processing system and the fault diagnosis system. During image processing, if the data processing system detects dirt or other issues in the acquired image data, it transmits the relevant signals to the data quality monitoring system. Similarly, if the fault diagnosis system finds that the quality of the transmitted images does not meet requirements, it also sends a relevant signal to the data quality monitoring system. Upon receiving a cleaning request from either the data processing system or the fault diagnosis system, the data quality monitoring system activates the corresponding cleaning device (such as a nozzle) to clean the lens and dust cover. After cleaning, it sends a relevant signal to the data acquisition system to reacquire the relevant data and repeat the fault diagnosis process.

[0049] (5) Alarm control system When the fault diagnosis system determines that there is an abnormality in the roller pins after diagnosis and analysis, it transmits the relevant fault information to the alarm control system. The alarm control system sends an alarm signal to the host computer through the control module and links the data acquisition system to perform re-acquisition and image reconstruction in order to verify and confirm the fault status and reduce false alarms generated by the system.

[0050] like Figure 3 As shown, this embodiment also provides a method for diagnosing pins on the roller surface of a high-pressure roller mill by fusing vibration and images. The diagnostic method includes the following steps S1-S4.

[0051] S1. After the system starts, it performs initial configuration. By reading the operating parameters of each module of vibration detection and image detection, such as sampling frequency, image resolution, alarm threshold, etc., it ensures the standardization of data acquisition and analysis. The vibration signals at both ends of the roller and the image data of the roller surface are collected through the data acquisition system.

[0052] S2. The vibration signal is denoised using the Time-Domain Synchronous Averaging (TSA) algorithm through the data processing system. The sample data is then processed using frequency domain analysis algorithms (Fourier transform, etc.) to extract important frequency domain features related to vibration, including margin factor, kurtosis, and root mean square frequency. The results are then input into the vibration anomaly detection model based on Support Vector Data Description (SVDD), which has been trained offline, to obtain the detection results of four measurement points at both ends of the moving and fixed rollers. Using the DS evidence fusion theory, the conflict factor of the diagnostic results at both ends of the rollers is first calculated. Then, based on the conflict factor, the completely contradictory parts of the diagnostic results at both ends are excluded (these parts of the diagnostic results are regarded as diagnostic anomalies caused by interference). The remaining diagnostic results are then fused to obtain the vibration anomaly detection report of the moving and fixed rollers.

[0053] S3. If the vibration analysis module detects an anomaly on the roller surface, the image recognition module is used for further verification. After the data processing system performs denoising, correction and data enhancement on the collected roller surface image, it is input into the image anomaly detection model based on deep learning network, and the output is the fault type (pin breakage, missing pin, wear, etc.) and spatial positioning result of the roller surface pins.

[0054] S4. The fault diagnosis system performs a comprehensive analysis based on the detection results of the vibration analysis module and the image recognition module. If a fault is diagnosed in the equipment, the alarm control system is triggered to send an audible and visual alarm to the host computer and link the data acquisition system to re-acquire and verify the data, thereby confirming the fault status.

[0055] Through the above diagnostic process, this invention can achieve collaborative judgment of vibration and image, rapid response to anomalies and adaptive adjustment, thereby forming a closed-loop, intelligent and dynamically evolving fault identification and processing mechanism, providing strong support for industrial-grade intelligent maintenance.

[0056] This invention integrates vibration detection and image detection. Due to excessive interference with vibration signals and limited acquisition channels, the vibration analysis module can only determine whether the roller has a fault, but it is difficult to pinpoint the fault. However, this module is efficient, fast, and has strong immediacy. The image recognition module can acquire a large amount of effective information through industrial cameras and accurately determine the fault type and location, but it requires a large amount of computation and has relatively low diagnostic efficiency. This invention organically combines the two, firstly using vibration signals to discover fault information instantly and efficiently, and then using image data to identify the fault type and locate the fault location, thus achieving efficient, accurate, and real-time fault detection.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A high-pressure roller mill roller surface pin diagnostic system based on vibration and image fusion, comprising a data acquisition system, a data processing system, a fault diagnosis system, and an alarm control system, characterized in that, The data acquisition system includes an image acquisition module and a vibration acquisition module. The image acquisition module acquires images of the roller surface using an industrial camera installed near the roller surface, while the vibration acquisition module acquires vibration signals using acceleration vibration sensors arranged near the bearing seats at both ends of the roller. The data processing system performs multi-dimensional preprocessing and feature extraction on the image data and vibration signals. The fault diagnosis system includes a vibration analysis module and an image recognition module. The vibration analysis module constructs a vibration anomaly detection model based on Support Vector Data Description (SVDD) based on the extracted vibration signal features to identify structural anomalies caused by early, minor changes. The image recognition module uses an image anomaly detection model based on a deep learning network to achieve accurate identification and spatial positioning of structural problems on the roller surface. When the fault diagnosis system detects a fault in the equipment, it transmits the relevant fault data to the alarm control system. The alarm control system sends an alarm signal to the host computer through the control module and links the image acquisition system to perform re-acquisition and image reconstruction to confirm the fault status.

2. The high-pressure roller mill roller surface pin diagnosis system based on vibration and image fusion according to claim 1, characterized in that, The diagnostic system also includes a data quality monitoring system. The data quality monitoring system uses a built-in image sharpness evaluation algorithm to monitor the quality indicators of each frame of image in real time and dynamically compares them with the working threshold of the lens transmittance factor set by the system. If the image is found to be blurry or the transmittance drops beyond the preset working threshold, it can automatically determine that there is dust, dirt or other obstructions on the lens surface and trigger a cleaning mechanism. After cleaning, the system automatically performs a data quality re-check and confirms whether the cleaning effect meets the standard before deciding whether to re-acquire image data.

3. A method for diagnosing pins on the roller surface of a high-pressure roller mill by vibration and image fusion, characterized in that, This diagnostic method uses the diagnostic system described in any one of claims 1-2 to diagnose roller surface pin faults, and specifically includes the following steps: S1. System parameter initialization: The vibration signals and roller surface image data at both ends of the roller are collected through the data acquisition system. S2. The vibration signal is denoised by applying a time-domain synchronous averaging algorithm to the data processing system, and frequency domain features are extracted by a frequency domain analysis algorithm. Then, the signals are input into the pre-trained vibration anomaly detection model based on support vector data description (SVDD) in the vibration analysis module for detection. The detection results of four measuring points at both ends of the moving roll and the fixed roll are obtained. Then, the DS evidence fusion theory is used to fuse the detection results of the four measuring points to obtain the vibration anomaly detection report of the moving roll and the fixed roll. S3. If the vibration analysis module detects an abnormality on the roller surface, the image recognition module is used for further verification. After the data processing system preprocesses the collected roller surface image, it is input into the image anomaly detection model based on a deep learning network, and the fault type and spatial location result of the roller surface pin are output. S4. If the fault diagnosis system detects a fault in the equipment, it will trigger the alarm control system to send an audible and visual alarm to the host computer and link the data acquisition system to re-acquire and verify the data, thereby confirming the fault status.

4. The method for diagnosing high-pressure roller mill roller surface pins by vibration and image fusion according to claim 3, characterized in that, In step S2, the frequency domain features extracted by the frequency domain analysis algorithm include margin factor, kurtosis and root mean square frequency.

5. The method for diagnosing high-pressure roller mill roller surface pins by vibration and image fusion according to claim 3, characterized in that, In step S2, the DS evidence fusion theory calculates the conflict factor between the diagnostic results at both ends and excludes the completely conflicting diagnostic results as "interference between the gearbox and the motor", thereby obtaining a more accurate diagnostic result.

6. The method for diagnosing high-pressure roller mill roller surface pins by vibration and image fusion according to claim 3, characterized in that, In step S2, the vibration signal is denoised using a time-domain synchronous averaging algorithm. The specific formula is as follows: ； In the formula, y(n) is the original signal. Let N be the denoised signal, N be the number of averages, k be the summation index, L be the length of the truncated window, and T be the window sliding step size. Both L and T are integer multiples of the rotation period.

7. The method for diagnosing high-pressure roller mill roller surface pins by vibration and image fusion according to claim 3, characterized in that, In step S3, the data processing system characterizes the image stability changes caused by abnormal roller surface vibration during image acquisition using a jitter index. The jitter index is determined by extracting displacement differences from consecutive image frames. The formula for the jitter index is: ; In the formula, The average displacement, i.e., the degree of chattering. To monitor the number of frames within a period, For the first The displacement difference between the first frame and the next frame.

8. The method for diagnosing high-pressure roller mill roller surface pins by vibration and image fusion according to claim 7, characterized in that, The fault diagnosis system calculates image jitter index to achieve image quality assessment and adaptive filtering. When assessing image quality, a threshold is set. When the jitter index exceeds the preset threshold, it indicates that there is significant jitter in the current image sequence, and the image is deemed unusable and re-acquired. When the jitter index is below the threshold, it indicates that the image acquisition process is stable and the quality is qualified.

9. The method for diagnosing high-pressure roller mill roller surface pins by vibration and image fusion according to claim 3, characterized in that, During image data acquisition, the image sharpness evaluation algorithm built into the data quality monitoring system monitors the quality index of each frame of the acquired image in real time and dynamically compares it with the working threshold of the lens transmittance factor set by the system. If the image quality index is lower than the preset working threshold, it can be automatically determined that there is dirt on the lens surface of the industrial camera and trigger the cleaning mechanism to clean the lens. After cleaning is completed, the image quality index is recalculated. If the cleaning effect is confirmed to meet the standard, image acquisition is resumed.

10. The method for diagnosing high-pressure roller mill roller surface pins by vibration and image fusion according to claim 9, characterized in that, The working threshold for the lens transmittance factor is established using the following formula: In the formula, The working threshold, This is the time interval since the last dust removal. The preset dust removal time interval, For time scale parameters, Transmittance measured in real time. Transmittance is for reference or ideal conditions. To detect comprehensive dust indicators from multiple angles, The minimum amount of dust to begin to attract attention. The scaling parameters for the air soot blowing response curve. , , These are the weighting coefficients for the timed dust removal factor, the light transmittance factor, and the multi-angle air blowing factor, respectively, and the sum of their coefficients is 1.

Images