Method for predicting case fault based on vibration data

Abstract

The invention discloses a method for predicting a case fault based on vibration data, and the method comprises the steps: collecting an original vibration acceleration signal of each to-be-monitored mechanical part in a working state in real time at a preset sampling frequency through vibration sensors which are arranged on each to-be-monitored mechanical part of a to-be-monitored case in advance; performing preprocessing including denoising and normalization processing on the collected original vibration acceleration signal of each to-be-monitored mechanical part in the working state to obtain a preprocessed vibration acceleration signal of each to-be-monitored mechanical part; performing frequency domain feature extraction on the preprocessed vibration acceleration signal of each to-be-monitored mechanical part to obtain a frequency domain feature vector corresponding to each to-be-monitored mechanical part; and inputting the constructed frequency domain feature vector into a corresponding trained machine learning prediction model, and performing mode comparison through a built-in fault fingerprint database of the model to obtain a scoring result of the health state of the corresponding to-be-monitored mechanical part.

Description

Technical Field

[0001] This invention relates to the field of technology, and more specifically, to a method for predicting chassis failures based on vibration data. Background Technology

[0002] In the daily operation of data centers, cloud computing clusters, and large server rooms, the chassis, as the carrier and protector of the core components of the server, directly determines whether the overall performance of the server can be normalized and is related to the continuous availability of business systems. The current mainstream chassis system integrates a large number of interconnected mechanical and electronic components—from cooling fans that ensure heat dissipation and hard drives that store core data, to brackets that fix the CPU heatsink, and the frame that supports the entire chassis structure. During long-term, continuous high-load operation, these components inevitably experience various potential failures due to wear, aging, or external forces. These include loose screws at mechanical connection points, imbalances in fan blades due to dust accumulation or slight deformation, wear on bearings after prolonged high-speed rotation, and resonance generated by the bracket under specific operating conditions.

[0003] To ensure stable operation of the server rack, current maintenance work mainly relies on regular manual inspections and three traditional maintenance methods. Firstly, regular manual inspections are the most basic maintenance method. Maintenance engineers need to go to the server room at fixed intervals to listen to the noise of the equipment during operation and feel the vibration intensity by touching the server rack to determine if there are any abnormalities in the internal components. However, this method has the following drawbacks: Firstly, the inspection frequency is limited by manpower and maintenance plans, making it difficult to achieve 24 / 7 real-time monitoring. If a fault begins to appear between two inspections, it is easily missed. Secondly, the judgment results are highly dependent on the engineer's experience. For example, the definition of abnormal noise or slight vibration lacks a unified standard and is highly subjective. Even experienced engineers find it difficult to accurately identify early, subtle signs of faults. Secondly, some mid-to-high-end servers employ temperature and fan speed-based health management solutions. These solutions monitor internal chassis temperature changes and real-time fan speeds using sensors. Alarms are triggered when these indicators exceed preset thresholds. However, in practice, this monitoring logic is severely inadequate for early identification of mechanical faults. For example, minor fan blade imbalances causing localized vibrations or slight loosening of bracket connecting screws do not directly affect fan speed or quickly cause abnormal internal chassis temperatures. Often, the system only issues an alarm when the fault has progressed to a certain extent, such as increased fan imbalance leading to speed fluctuations or loose brackets generating additional heat through friction. By this time, the fault is nearing a critical stage, significantly increasing the difficulty of repair and the risk of downtime. Thirdly, a few solutions attempt to diagnose faults through sound monitoring. This involves using microphones to collect sound signals from the chassis during operation and then using spectrum analysis to identify anomalies. However, in data center environments, dozens or even hundreds of devices typically operate simultaneously, and the combined noise from various devices creates complex background interference. In this situation, the signal collected by the microphone is easily drowned out by environmental noise. Even with algorithmic processing, it is difficult to accurately extract the fault sound characteristics of a single chassis. As a result, the solution has extremely poor robustness in data centers with dense deployment of multiple devices and has limited practical value.

[0004] Therefore, there is an urgent need in the industry for a monitoring solution that can be designed to detect early signs of chassis mechanical failures. Summary of the Invention

[0005] This specification provides a method for predicting chassis failures based on vibration data, in order to overcome at least one technical problem existing in related technologies.

[0006] According to embodiments of this specification, a method for predicting chassis failures based on vibration data is provided, including: Vibration sensors are pre-placed on each mechanical component of the chassis to be monitored, and the original vibration acceleration signal of each mechanical component to be monitored when it is in working state is collected in real time at a preset sampling frequency. The original vibration acceleration signals of each monitored mechanical component under working conditions are preprocessed, including noise reduction and normalization, to obtain the preprocessed vibration acceleration signals of each monitored mechanical component. Frequency domain feature extraction is performed on the preprocessed vibration acceleration signal of each mechanical component to be monitored to obtain the frequency domain feature vector corresponding to each mechanical component to be monitored. The constructed frequency domain feature vector is input into the corresponding trained machine learning prediction model. The model's built-in fault fingerprint database is used for pattern comparison to obtain the health status score of the corresponding mechanical component to be monitored. The fault fingerprint database stores the spectral patterns corresponding to different types of mechanical faults. The trained machine learning prediction model is trained based on historical vibration data and known fault type labels, and outputs a score to characterize the health status of the corresponding mechanical component to be monitored. When the health status score of the corresponding monitored mechanical component is lower than a preset threshold, an early warning signal is generated and an alarm message is sent to the operation and maintenance terminal through at least one communication protocol.

[0007] In some alternative implementations, the vibration sensor is a triaxial accelerometer, and the mechanical component to be monitored includes the fan mounting area, bracket connection area, and hard drive tray mounting area of ​​the chassis.

[0008] In some alternative implementations, the preset sampling frequency is adjustable in the range of 5 kHz to 20 kHz and is configured according to the inherent vibration characteristics of the mechanical component being monitored.

[0009] In some optional implementations, the preprocessing of the raw vibration acceleration signals of each monitored mechanical component in its working state, including noise reduction and normalization, includes: The wavelet transform algorithm is used to denoise the acquired raw vibration acceleration signal in order to filter out the temperature drift interference components with frequencies lower than the predetermined frequency. The denoised signal is segmented into segments with a fixed time window of 1 second to obtain a continuous multi-segment signal sequence. The normalization process for each signal sequence is calculated using the following formula: Among them, symbols Represents the normalized signal value, symbol In a segmented signal sequence, the first segment represents the... The original value of each point, symbol This represents the mean of all points in the signal sequence, with the sign... It represents the standard deviation of the signal sequence.

[0010] In some optional implementations, the step of extracting frequency domain features from the preprocessed vibration acceleration signal of each monitored mechanical component to obtain a frequency domain feature vector corresponding to each monitored mechanical component includes: A fast Fourier transform is performed on the preprocessed vibration acceleration signal segment to convert the signal from the time domain to the frequency domain and obtain the corresponding spectrum data. Identify and record the peak frequencies corresponding to the five highest amplitudes in the spectrum and their corresponding amplitudes; Extract the amplitude of the harmonic components related to the fundamental frequency of component rotation, including calculating the amplitude at the second and third harmonics; Calculate the total energy of the vibration signal in the frequency band from 50Hz to 2000Hz, and characterize it as the root mean square value; Analyze the energy distribution characteristics of the spectrum and calculate its skewness and kurtosis; The extracted peak frequency, peak amplitude, octave amplitude, root mean square value, skewness, and kurtosis are combined to form a multidimensional frequency domain feature vector that characterizes the current vibration state of the mechanical component.

[0011] In some alternative implementations, the warning signal is sent to the maintenance personnel's maintenance terminal via at least one of the following methods: SNMP protocol, email, or SMS.

[0012] In some optional implementations, the system also includes displaying the current vibration spectrum, model prediction results, and remaining service life estimate calculated based on the prediction results in a graphical user interface in real time.

[0013] In some alternative implementations, the trained machine learning prediction model employs a gradient boosting tree or a one-dimensional convolutional neural network model.

[0014] According to a second aspect of the present invention, an apparatus for predicting chassis failures based on vibration data is also provided, the apparatus comprising: The vibration data acquisition module is used to acquire the original vibration acceleration signal of each mechanical component under monitoring when it is in working state in real time by means of vibration sensors pre-arranged on each mechanical component under monitoring in the chassis to be monitored, at a preset sampling frequency. The data preprocessing module is used to preprocess the original vibration acceleration signal of each monitored mechanical component when it is in working state, including noise reduction and normalization, to obtain the preprocessed vibration acceleration signal of each monitored mechanical component. The frequency domain feature extraction module is used to extract frequency domain features from the preprocessed vibration acceleration signal of each mechanical component to be monitored, and obtain the frequency domain feature vector corresponding to each mechanical component to be monitored. The prediction model analysis module is used to input the constructed frequency domain feature vector into the corresponding trained machine learning prediction model, and to obtain the health status score of the corresponding mechanical component to be monitored by comparing patterns through the fault fingerprint database built into the model. The fault fingerprint database stores the spectral patterns corresponding to different types of mechanical faults, and the trained machine learning prediction model is trained based on historical vibration data and known fault type labels, and outputs the score to characterize the health status of the corresponding mechanical component to be monitored. The alarm generation and sending module is used to generate an early warning signal and send alarm information to the operation and maintenance terminal through at least one communication protocol when the health status score of the corresponding monitored mechanical component is lower than a preset threshold.

[0015] The beneficial effects of the embodiments in this specification are as follows: This technical solution acquires raw vibration acceleration signals of each mechanical component under monitoring in real time. After preprocessing and frequency domain feature extraction, the frequency domain feature vector is input into the corresponding trained machine learning prediction model. The model is then compared with a fault fingerprint database containing spectral patterns of different types of mechanical faults to identify early, minute vibration anomalies in the mechanical components of the chassis. This avoids the problem of existing chassis monitoring systems failing to detect early mechanical faults in a timely manner. It can predict faults before they develop into severe stages, thereby preventing equipment downtime caused by untimely fault handling, reducing business interruptions caused by single chassis faults affecting adjacent servers, and lowering the maintenance costs and risks associated with emergency repairs. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments or related technologies of this specification, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a method for predicting chassis failures based on vibration data, provided in this application; Figure 2 The corresponding to this application Figure 1 A structural diagram of a device for predicting chassis failures based on vibration data. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0019] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0020] In the description of this embodiment, the terms "upper," "lower," "right," etc., refer to the orientation or positional relationship shown in the accompanying drawings. They are used only for ease of description and simplification of operation, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention. In addition, the terms "first" and "second" are used only for distinction in description and have no special meaning.

[0021] Addressing the shortcomings of existing technologies mentioned in the background section, the applicant has observed through long-term practice that common mechanical failures of computer cases, such as fan imbalance, loose brackets, or vibrations caused by unstable hard drive trays, are accompanied by specific vibration pattern changes in their early stages. These changes exhibit distinct characteristics in the frequency domain (i.e., the correspondence between frequency and vibration amplitude). For example, when a fan is imbalanced, abnormal vibration amplitude peaks appear near its own rotational speed multiples, while loose brackets lead to a significant increase in vibration amplitude within a specific frequency range. Based on this consideration, the applicant designed the technical solution presented in this application. This solution, by real-time acquisition of vibration signals from the monitored components, can accurately extract frequency domain features from the vibration data and combine this with a predictive model trained on historical failure samples to achieve early warning of failures. This transforms the operation and maintenance mode from passive repair to proactive prevention, ultimately improving the reliability of data center equipment operation and reducing losses caused by failures.

[0022] The technical solution of this application will be described below with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for predicting chassis faults based on vibration data, as provided in an embodiment of this specification. The following is based on... Figure 1The technical solution of this application is described, and the method may include the following steps: Step 102: Using vibration sensors pre-placed on each mechanical component to be monitored in the chassis to be monitored, the original vibration acceleration signal of each mechanical component to be monitored when it is in working state is collected in real time at a preset sampling frequency.

[0023] In this step, vibration sensors are pre-positioned on each mechanical component of the chassis to be monitored. Three-axis accelerometers can be used as the vibration sensors. When selecting which specific mechanical components to monitor, priority should be given to components prone to mechanical failure due to their structural characteristics or operating modes during long-term operation. For example, the fan in the chassis needs to rotate at high speed, and it is prone to imbalance due to dust accumulation / deformation of the blades, and bearing wear. Similarly, the hard drive tray needs to buffer hard drive vibration, and it is prone to abnormal vibration transmission due to loose fixing screws or tray deformation. Furthermore, the bracket needs to bear loads and be fixed for a long time, and it is prone to unstable connections due to loose screws or structural aging. Therefore, in a real-world scenario, these mechanical components can be listed as the monitoring targets.

[0024] Each monitored mechanical component can be equipped with at least one triaxial accelerometer to directly capture the vibration signal of a single component, avoiding interference between vibration signals from different components. Simultaneously, the original vibration acceleration signal of each monitored mechanical component in its operating state is acquired in real time at a preset sampling frequency. This preset sampling frequency can be in the range of 5kHz to 20kHz and can be adjusted according to the operating frequency characteristics of the monitored mechanical component. During the acquisition process, the triaxial accelerometer captures the original analog vibration acceleration signal generated by the corresponding monitored mechanical component in real time. The acquired original analog vibration acceleration signal can be converted into a digital form of original vibration acceleration signal using a high-speed analog-to-digital converter, thereby obtaining the original vibration acceleration signal of each monitored mechanical component in its operating state.

[0025] Step 104: Perform preprocessing, including denoising and normalization, on the original vibration acceleration signal of each monitored mechanical component when it is in working state to obtain the preprocessed vibration acceleration signal of each monitored mechanical component.

[0026] In this step, the raw vibration acceleration signals of each monitored mechanical component under operating conditions undergo preprocessing, including denoising and normalization. Specifically, wavelet transform algorithms are first used to denoise these raw vibration acceleration signals to filter out temperature drift interference components with frequencies below a predetermined frequency, such as 5Hz, leaving signals that accurately reflect the vibration state of the component. Next, the denoised signals are segmented into fixed time windows of 1 second, breaking the continuous signal into multiple consecutive signal sequences. This segmentation allows for more targeted signal processing and facilitates feature extraction in fixed units. Finally, normalization is performed on each signal sequence to obtain the preprocessed vibration acceleration signal corresponding to each monitored mechanical component.

[0027] Step 106: Extract frequency domain features from the preprocessed vibration acceleration signal of each mechanical component to be monitored, and obtain the frequency domain feature vector corresponding to each mechanical component to be monitored.

[0028] In this step, frequency domain features are extracted from the preprocessed vibration acceleration signal of each monitored mechanical component. Specifically, the preprocessed vibration acceleration signal can be converted from the time domain to the frequency domain using either Fast Fourier Transform (FFT) or Short-Time Fourier Transform (SFT). Using FFT yields the corresponding signal's spectral data, while using SFT allows for processing with a preset window function to obtain a two-dimensional time-frequency matrix. Features reflecting the component's vibration characteristics are then extracted from the converted frequency domain data. This involves identifying the peak frequency and its corresponding peak amplitude, calculating the overall energy of the vibration signal within a specific frequency band, and characterizing it using the root mean square (RMS) value. Simultaneously, the energy distribution characteristics of the spectrum are analyzed to calculate skewness and kurtosis.

[0029] Considering that the technical solution of this application requires monitoring the health status of different types of mechanical components, and that the operating principles and fault manifestations of different types of mechanical components may differ, appropriate frequency domain feature vectors can be set according to the properties of different mechanical components being monitored when constructing frequency domain feature vectors. For example, the function of a bracket is to fix CPU heatsinks, chassis frames, etc., and its operating characteristics are static support and passive vibration bearing. The failure modes of this type of component are mainly screw loosening or structural resonance. Since this type of failure does not generate rotational harmonics, but rather causes a significant increase in the vibration amplitude of the bracket's own inherent resonant frequency range, and the kurtosis of the vibration signal will increase due to the impact vibration caused by loosening. Therefore, when extracting frequency domain features for components like brackets, we can disregard harmonic amplitude and harmonic ratio, and instead focus on features reflecting structural stability such as peak amplitude, kurtosis, and skewness in specific frequency ranges. For example, the function of a hard drive bracket is to secure the hard drive and buffer its vibrations during operation. The bracket generates low-frequency vibrations during hard drive read / write operations and must withstand the periodic vibrations of the hard drive. Therefore, hard drive bracket failure modes may primarily involve loose mounting screws or bracket deformation leading to abnormal vibration transmission. These failures will increase the root mean square (RMS) value of vibration near the hard drive's base frequency (because RMS reflects the average energy of the vibration signal; looseness will cause the vibration energy to disperse and increase overall), and the peak frequency may deviate from the normal hard drive base frequency range. Another example is the high-speed rotation of a fan. Failure modes are primarily blade imbalance and bearing wear. These failures will cause abnormal vibration amplitude peaks near integer multiples of the fan's base frequency (such as 1x, 2x, 3x). Therefore, if the mechanical component to be monitored is a component with a rotating fundamental frequency, such as a chassis fan, the first harmonic amplitude, second harmonic amplitude, and third harmonic amplitude corresponding to its rotating fundamental frequency can be extracted. Finally, all these extracted features are arranged in a preset fixed order to form a frequency domain feature vector that corresponds only to this mechanical component to be monitored and can accurately characterize its current vibration state.

[0030] Step 108: Input the constructed frequency domain feature vector into the corresponding trained machine learning prediction model, and perform pattern comparison through the fault fingerprint database built into the model to obtain the health status score of the corresponding mechanical component to be monitored; wherein, the fault fingerprint database stores the spectral patterns corresponding to different types of mechanical faults, and the trained machine learning prediction model is trained based on historical vibration data and known fault type labels, and outputs the score to characterize the health status of the corresponding mechanical component to be monitored.

[0031] In this step, the constructed frequency domain feature vector is input into the corresponding trained machine learning prediction model. The model contains a fault fingerprint database, which stores spectral patterns corresponding to different types of mechanical faults, such as the spectral patterns of common chassis mechanical faults like fan imbalance, bracket loosening, and hard drive tray vibration. During model operation, the input frequency domain feature vector is compared with each fault spectral pattern in the fault fingerprint database. By analyzing the similarity between the current feature vector and different fault patterns, the component's condition is determined. Finally, based on the pattern comparison results and pre-learned correlations, the model outputs a score that accurately represents the health status of the corresponding monitored mechanical component. The trained machine learning prediction model can employ a gradient boosting tree or a one-dimensional convolutional neural network model.

[0032] Step 110: When the health status score of the corresponding mechanical component to be monitored is lower than a preset threshold, an early warning signal is generated and an alarm message is sent to the operation and maintenance terminal through at least one communication protocol.

[0033] It should be noted that step 106 above explained in detail why it is necessary to set appropriate frequency domain feature vectors according to the properties of different mechanical parts to be monitored when constructing frequency domain feature vectors. Based on this, the number of machine learning prediction models trained in this step is also multiple. That is, considering that the vibration characteristics and fault modes of different parts are not completely different, it is necessary to train a dedicated prediction model for each mechanical part to be monitored. Each model only inputs the frequency domain feature vector of the corresponding part and only matches the fault fingerprint database of that part, thereby ensuring the exclusive correlation between features and fault modes.

[0034] This technical solution acquires raw vibration acceleration signals of each mechanical component under monitoring in real time. After preprocessing and frequency domain feature extraction, the frequency domain feature vector is input into the corresponding trained machine learning prediction model. The model is then compared with a fault fingerprint database containing spectral patterns of different types of mechanical faults to identify early, minute vibration anomalies in the mechanical components of the chassis. This avoids the problem of existing chassis monitoring systems failing to detect early mechanical faults in a timely manner. It can predict faults before they develop into severe stages, thereby preventing equipment downtime caused by untimely fault handling, reducing business interruptions caused by single chassis faults affecting adjacent servers, and lowering the maintenance costs and risks associated with emergency repairs.

[0035] Based on the technical solutions described above, this application also provides some more specific technical solutions, which are described below.

[0036] In an optional embodiment, the vibration sensor may be a triaxial accelerometer, and the mechanical component to be monitored may include the fan mounting location, bracket connection location, and hard drive tray mounting location of the chassis.

[0037] In an optional embodiment, the preset sampling frequency is adjustable in the range of 5kHz to 20kHz and is configured according to the inherent vibration characteristics of the mechanical component being monitored.

[0038] In this embodiment, considering that different mechanical components to be monitored (such as the fan, bracket, and hard drive tray of the chassis mentioned in the previous embodiment) have different inherent vibration frequencies due to differences in structure, material, and operation mode, that is, when the fan rotates at high speed, the blade vibration and bearing operation are prone to generate higher frequency vibration signals, while if the bracket connection is loose, the vibration frequency is usually relatively low. Therefore, different sampling frequencies need to be set for different mechanical components to be monitored.

[0039] In an optional embodiment, the preprocessing of the original vibration acceleration signal of each monitored mechanical component in its working state, including noise reduction and normalization, includes: The wavelet transform algorithm is used to denoise the acquired raw vibration acceleration signal in order to filter out the temperature drift interference components with frequencies lower than the predetermined frequency. The denoised signal is segmented into segments with a fixed time window of 1 second to obtain a continuous multi-segment signal sequence. The normalization process for each signal sequence is calculated using the following formula: Among them, symbols Represents the normalized signal value, symbol In a segmented signal sequence, the first segment represents the... The original value of each point, symbol This represents the mean of all points in the signal sequence, with the sign... It represents the standard deviation of the signal sequence.

[0040] The technical solution of this embodiment mainly aims to preprocess the acquired raw vibration acceleration signal. First, denoising is performed, specifically using a wavelet transform algorithm to operate on the raw vibration acceleration signal. Because wavelet transform has the characteristic of multi-resolution analysis in the time and frequency domain, it can accurately identify and filter out temperature drift interference components with frequencies below a predetermined frequency, such as 5Hz. This type of interference is mainly caused by environmental temperature fluctuations, not by the vibration of mechanical components, and will cause the raw signal to be mixed with false fluctuations, interfering with the accuracy of subsequent analysis. In this embodiment, the denoising process using wavelet transform first effectively retains the valid signal components reflecting the mechanical vibration state.

[0041] Subsequently, the denoised signal is segmented into segments with a fixed time window of 1 second. The technical solution in this embodiment selects a time scale of 1 second, which can capture the real-time vibration characteristics of mechanical parts and ensure that each segment of the signal contains a sufficient amount of data.

[0042] Finally, based on the above formula, normalization is performed on each signal sequence. In this embodiment, normalization can eliminate the influence of amplitude differences in signals at different time periods, so that the signals are presented under a unified dimension.

[0043] In an optional embodiment, the step of extracting frequency domain features from the preprocessed vibration acceleration signal of each monitored mechanical component to obtain a frequency domain feature vector corresponding to each monitored mechanical component may include: A fast Fourier transform is performed on the preprocessed vibration acceleration signal segment to convert the signal from the time domain to the frequency domain and obtain the corresponding spectrum data. Identify and record the peak frequencies corresponding to the five highest amplitudes in the spectrum and their corresponding amplitudes; Extract the amplitude of the harmonic components related to the fundamental frequency of component rotation, including calculating the amplitude at the second and third harmonics; Calculate the total energy of the vibration signal in the frequency band from 50Hz to 2000Hz, and characterize it as the root mean square value; Analyze the energy distribution characteristics of the spectrum and calculate its skewness and kurtosis; The extracted peak frequency, peak amplitude, octave amplitude, root mean square value, skewness, and kurtosis are combined to form a multidimensional frequency domain feature vector that characterizes the current vibration state of the mechanical component.

[0044] In an optional embodiment, the warning signal is sent to the maintenance personnel's terminal via at least one of the following methods: SNMP protocol, email, or SMS.

[0045] In optional embodiments, the technical solution also includes displaying the current vibration spectrum, model prediction results, and remaining service life estimate calculated based on the prediction results in a graphical user interface in real time.

[0046] It should be understood that in the methods described in one or more embodiments of this specification, the order of some steps may be adjusted according to actual needs, or some steps may be omitted.

[0047] Based on the same idea, embodiments of this specification also provide apparatus corresponding to the above methods. Figure 2 The embodiments provided in this specification correspond to Figure 1 A schematic diagram of a device for predicting chassis failures based on vibration data. Figure 2 As shown, the device may include: The vibration data acquisition module 202 is used to acquire the original vibration acceleration signal of each of the mechanical components under monitoring when it is in working state in real time by means of vibration sensors pre-arranged on each mechanical component under monitoring in the chassis to be monitored, at a preset sampling frequency. The data preprocessing module 204 is used to preprocess the original vibration acceleration signal of each monitored mechanical component when it is in working state, including noise reduction and normalization, to obtain the preprocessed vibration acceleration signal of each monitored mechanical component. The frequency domain feature extraction module 206 is used to extract frequency domain features from the preprocessed vibration acceleration signal of each mechanical component to be monitored, and obtain the frequency domain feature vector corresponding to each mechanical component to be monitored. The prediction model analysis module 208 is used to input the constructed frequency domain feature vector into the corresponding trained machine learning prediction model, and to obtain the health status score of the corresponding mechanical component to be monitored by comparing patterns through the fault fingerprint database built into the model; wherein, the fault fingerprint database stores the spectral patterns corresponding to different types of mechanical faults, and the trained machine learning prediction model is trained based on historical vibration data and known fault type labels, and outputs the score to characterize the health status of the corresponding mechanical component to be monitored. The alarm generation and sending module 210 is used to generate an alarm signal and send alarm information to the operation and maintenance terminal through at least one communication protocol when the health status score of the corresponding monitored mechanical component is lower than a preset threshold.

[0048] It is understood that the modules mentioned above refer to computer programs or program segments used to perform one or more specific functions. Furthermore, the distinction between these modules does not imply that the actual program code must also be separate.

[0049] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art will be able to make various obvious changes, readjustments, and substitutions without departing from the scope of protection of the present invention. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting chassis failures based on vibration data, characterized in that, Includes the following steps: Vibration sensors are pre-placed on each mechanical component of the chassis to be monitored, and the original vibration acceleration signal of each mechanical component to be monitored when it is in working state is collected in real time at a preset sampling frequency. The original vibration acceleration signals of each monitored mechanical component under working conditions are preprocessed, including noise reduction and normalization, to obtain the preprocessed vibration acceleration signals of each monitored mechanical component. Frequency domain feature extraction is performed on the preprocessed vibration acceleration signal of each mechanical component to be monitored to obtain the frequency domain feature vector corresponding to each mechanical component to be monitored. The constructed frequency domain feature vector is input into the corresponding trained machine learning prediction model. The model's built-in fault fingerprint database is used for pattern comparison to obtain the health status score of the corresponding mechanical component to be monitored. The fault fingerprint database stores the spectral patterns corresponding to different types of mechanical faults. The trained machine learning prediction model is trained based on historical vibration data and known fault type labels, and outputs a score to characterize the health status of the corresponding mechanical component to be monitored. When the health status score of the corresponding monitored mechanical component is lower than a preset threshold, an early warning signal is generated and an alarm message is sent to the operation and maintenance terminal through at least one communication protocol.

2. The method for predicting chassis failures based on vibration data according to claim 1, characterized in that, The vibration sensor is a triaxial accelerometer, and the mechanical components to be monitored include the fan mounting area, bracket connection area, and hard drive tray mounting area of ​​the chassis.

3. The method for predicting chassis failures based on vibration data according to claim 1, characterized in that, The preset sampling frequency is adjustable in the range of 5kHz to 20kHz and is configured according to the inherent vibration characteristics of the mechanical component being monitored.

4. The method for predicting chassis failures based on vibration data according to claim 1, characterized in that, The preprocessing of the original vibration acceleration signals of each monitored mechanical component in its working state, including noise reduction and normalization, includes: The wavelet transform algorithm is used to denoise the acquired raw vibration acceleration signal in order to filter out the temperature drift interference components with frequencies lower than the predetermined frequency. The denoised signal is segmented into segments with a fixed time window of 1 second to obtain a continuous multi-segment signal sequence. The normalization process for each signal sequence is calculated using the following formula: Among them, symbols Represents the normalized signal value, symbol In a segmented signal sequence, the first segment represents the... The original value of each point, symbol This represents the mean of all points in the signal sequence, with the sign... It represents the standard deviation of the signal sequence.

5. The method for predicting chassis failures based on vibration data according to claim 4, characterized in that, The process of extracting frequency domain features from the preprocessed vibration acceleration signal of each monitored mechanical component to obtain the frequency domain feature vector corresponding to each monitored mechanical component includes: A fast Fourier transform is performed on the preprocessed vibration acceleration signal segment to convert the signal from the time domain to the frequency domain and obtain the corresponding spectrum data. Identify and record the peak frequencies corresponding to the five highest amplitudes in the spectrum and their corresponding amplitudes; Extract the amplitude of the harmonic components related to the fundamental frequency of component rotation, including calculating the amplitude at the second and third harmonics; Calculate the total energy of the vibration signal in the frequency band from 50Hz to 2000Hz, and characterize it as the root mean square value; Analyze the energy distribution characteristics of the spectrum and calculate its skewness and kurtosis; The extracted peak frequency, peak amplitude, octave amplitude, root mean square value, skewness, and kurtosis are combined to form a multidimensional frequency domain feature vector that characterizes the current vibration state of the mechanical component.

6. The method for predicting chassis failures based on vibration data according to claim 1, characterized in that, The warning signal is sent to the maintenance personnel's terminal via at least one of the following methods: SNMP protocol, email, or SMS.

7. The method for predicting chassis failures based on vibration data according to claim 6, characterized in that, It also includes real-time display of the current vibration spectrum, model prediction results, and remaining service life estimates calculated based on the prediction results in a graphical user interface.

8. The method for predicting chassis failures based on vibration data according to claim 1, characterized in that, The trained machine learning prediction model adopts either a gradient boosting tree or a one-dimensional convolutional neural network model.

9. A device for predicting chassis failures based on vibration data, characterized in that, The device includes: The vibration data acquisition module is used to acquire the original vibration acceleration signal of each mechanical component under monitoring when it is in working state in real time by means of vibration sensors pre-arranged on each mechanical component under monitoring in the chassis to be monitored, at a preset sampling frequency. The data preprocessing module is used to preprocess the original vibration acceleration signal of each monitored mechanical component when it is in working state, including noise reduction and normalization, to obtain the preprocessed vibration acceleration signal of each monitored mechanical component. The frequency domain feature extraction module is used to extract frequency domain features from the preprocessed vibration acceleration signal of each mechanical component to be monitored, and obtain the frequency domain feature vector corresponding to each mechanical component to be monitored. The prediction model analysis module is used to input the constructed frequency domain feature vector into the corresponding trained machine learning prediction model, and to obtain the health status score of the corresponding mechanical component to be monitored by comparing patterns through the fault fingerprint database built into the model. The fault fingerprint database stores the spectral patterns corresponding to different types of mechanical faults, and the trained machine learning prediction model is trained based on historical vibration data and known fault type labels, and outputs the score to characterize the health status of the corresponding mechanical component to be monitored. The alarm generation and sending module is used to generate an early warning signal and send alarm information to the operation and maintenance terminal through at least one communication protocol when the health status score of the corresponding monitored mechanical component is lower than a preset threshold.

Images