A method and related devices for early warning of disk array failures

Abstract

The present application discloses a method and related devices for disk array fault warning, which relates to the field of computer storage technology. The method includes: in response to a preset periodic trigger condition, applying a preset mechanical excitation signal to a target disk array; collecting the vibration response spectrum and acoustic response waveform of the target disk array under the action of the mechanical excitation signal; performing fusion analysis on the vibration response spectrum and acoustic response waveform through a preset feature fusion model to obtain a fault probability prediction vector; and determining a real-time fault warning instruction for the target disk array according to the fault probability prediction vector and a preset health benchmark vector. Through periodic mechanical excitation and multi-modal signal fusion analysis, the present application can achieve high-precision, non-invasive, and real-time fault warning for disk arrays, improve the reliability of disk arrays, and reduce operation and maintenance costs.

Description

Technical Field

[0001] This application relates to the field of computer storage technologies, and more specifically, to a method and related devices for early warning of disk array failures. Background Art

[0002] With the continuous expansion of information technology and the scale of data centers, as a core storage device, the operating stability of disk arrays has a crucial impact on the data security and business continuity of the entire system. However, disk arrays usually operate under complex working conditions such as high load, high density, and long-term operation, and are extremely vulnerable to factors such as mechanical wear, vibration shock, and poor heat dissipation, resulting in performance degradation or failures. In particular, hardware-level failures are difficult to detect in a timely manner, easily leading to serious consequences such as data loss and system downtime. Therefore, developing efficient and reliable fault early warning technologies has become a key means to ensure the stable operation of disk arrays.

[0003] In the prior art, most of the fault early warnings of disk arrays rely on SMART data monitoring, operation log analysis, or prediction methods based on statistical laws. Although the implementation is relatively simple, there are generally problems such as relying on internal indicators, lagging responses, and weak recognition capabilities for sudden physical failures, making it difficult to detect potential anomalies at the structural or hardware levels in a timely manner. In addition, traditional methods lack effective utilization of the external physical characteristics of disk arrays and cannot comprehensively evaluate their operating states from the perspective of mechanical behavior, resulting in insufficient accuracy and foresight of early warning results. That is, there are technical problems in the related technologies such as weak fault early warning capabilities at the physical level of disk arrays, lagging responses, single-dimensional monitoring data, and limited prediction accuracy. Summary of the Invention

[0004] A series of simplified concepts are introduced in the Summary of the Invention section of this application, which will be further elaborated in detail in the Detailed Implementation section. The Summary of the Invention section of this application does not mean to attempt to define the key features and essential technical features of the claimed technical solution, nor does it mean to attempt to determine the protection scope of the claimed technical solution.

[0005] The method and related devices for early warning of disk array failures provided by this application can achieve high-precision, non-invasive, and real-time fault early warning of disk arrays through periodic mechanical excitation and multi-modal signal fusion analysis, improving the reliability of disk arrays and reducing operation and maintenance costs.

[0006] In a first aspect, the present application provides a method for early warning of disk array failures, including: in response to a preset periodic trigger condition, applying a preset mechanical excitation signal to a target disk array; collecting a vibration response spectrum and an acoustic response waveform of the target disk array under the action of the mechanical excitation signal; performing fusion analysis on the vibration response spectrum and the acoustic response waveform through a preset feature fusion model to obtain a fault probability prediction vector; and determining a real-time fault early warning instruction for the target disk array according to the fault probability prediction vector and a preset health benchmark vector.

[0007] In some embodiments, the applying a preset mechanical excitation signal to a target disk array includes: generating a composite sweep signal having a first vibration frequency range and a second vibration frequency range, wherein the first vibration frequency range covers a disk bearing rotation characteristic frequency band of the target disk array, and the second vibration frequency range covers a head swing arm resonance frequency band of the target disk array; converting the composite sweep signal into the preset mechanical excitation signal through an embedded controller; and time-sharingly loading the preset mechanical excitation signal to a vibration exciter of the target disk array based on a preset time series.

[0008] In some embodiments, the collecting a vibration response spectrum and an acoustic response waveform of the target disk array under the action of the mechanical excitation signal includes: collecting an original vibration signal of the target disk array under the action of the mechanical excitation signal through a piezoelectric sensor array arranged on a disk bracket of the target disk array; collecting an original acoustic signal of the target disk array under the action of the mechanical excitation signal through a multi-channel microphone array arranged on the top of a disk bin of the target disk array; performing wavelet packet decomposition on the original vibration signal to obtain the vibration response spectrum; and performing mel cepstral coefficient transformation on the original acoustic signal to obtain the acoustic response waveform including time series characteristics.

[0009] In some embodiments, the preset feature fusion model includes: a first convolutional neural network branch for performing time-frequency feature extraction on the vibration response spectrum and outputting a first feature tensor; a second convolutional neural network branch for performing formant feature extraction on the acoustic response waveform and outputting a second feature tensor; and a dynamic attention fusion module for generating the fault probability feature vector according to the correlation weight between the first feature tensor and the second feature tensor.

[0010] In some embodiments, determining the real-time fault warning instruction for the target disk array according to the fault probability prediction vector and the preset health benchmark vector includes: determining the cosine similarity between the fault probability feature vector and the preset health benchmark vector; when the cosine similarity is less than the first threshold and greater than or equal to the second threshold, the real-time fault warning instruction is to trigger a primary warning and generate a disk health report, where the second threshold is less than the first threshold; when the cosine similarity is less than the second threshold and greater than or equal to the third threshold, the real-time fault warning instruction is to trigger an intermediate warning and start redundant data migration, where the third threshold is less than the second threshold; when the cosine similarity is less than the third threshold, the real-time fault warning instruction is to trigger a high-level warning and mark the physical location of the faulty disk.

[0011] In some embodiments, it further includes: generating physical location codes for each physical disk in the target disk array according to the preset physical topology structure of the target disk array; determining the fault probability values of the physical disks according to the sub-vector components corresponding to the physical disks in the target disk array in the fault probability prediction vector; determining the historical fault similarity of each physical disk in the target disk array according to the matching result between the vibration response spectrum and the historical fault case library.

[0012] In some embodiments, it further includes: displaying a three-dimensional fault hot spot distribution map in a graphical user interface, where the X-axis of the three-dimensional fault hot spot distribution map represents the physical location code, the Y-axis represents the fault probability value, and the Z-axis represents the historical fault similarity of each physical disk in the target disk array; determining the mapping color of the three-dimensional fault hot spot distribution map according to the warning level in the real-time fault warning instruction.

[0013] In a second aspect, the present application further provides a disk array fault warning device, including: a signal application unit for applying a preset mechanical excitation signal to the target disk array in response to a preset periodic trigger condition; a signal acquisition unit for acquiring the vibration response spectrum and the acoustic response waveform of the target disk array under the action of the mechanical excitation signal; a vector prediction unit for performing fusion analysis on the vibration response spectrum and the acoustic response waveform through a preset feature fusion model to obtain a fault probability prediction vector; an instruction determination unit for determining a real-time fault warning instruction for the target disk array according to the fault probability prediction vector and the preset health benchmark vector.

[0014] In a third aspect, the present application further provides an electronic device, including: a memory and a processor, where the processor is used to implement the steps of the disk array fault warning method described in the first aspect when executing the computer program stored in the memory.

[0015] In a fourth aspect, the present application further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the disk array fault warning method described in the first aspect are implemented.

[0016] In a fifth aspect, the present application further provides a computer program product including a computer program or computer-executable instructions, and when the computer program or computer-executable instructions are executed by a processor, the disk array fault warning method provided by the embodiments of the present application is implemented.

[0017] In summary, the present application can obtain the state information of the disk array from multiple dimensions by fusing the vibration response spectrum and the acoustic response waveform, overcomes the problem of insufficient prediction accuracy of a single signal source, and improves the accuracy of fault identification and prediction; uses an externally applied mechanical excitation instead of internal data monitoring, does not affect the normal operation of the disk array, and has strong engineering adaptability and safety; actively stimulates and collects response signals according to a preset period, which helps to identify potential anomalies earlier, improves the real-time performance and sensitivity of fault warning, and facilitates early intervention and maintenance; can issue a warning before the disk array has a serious fault, helps the operation and maintenance personnel to handle problems in time, thereby reducing the economic losses caused by system downtime, and improving the overall availability and reliability of the system. In summary, the disk array fault warning method provided by the present application can realize high-precision, non-intrusive, and real-time fault warning of the disk array through periodic mechanical excitation and multi-modal signal fusion analysis, improve the reliability of the disk array, and reduce the operation and maintenance cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to limit this specification. And throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0019] Figure 1 is a schematic flow chart of a disk array fault warning method provided by an embodiment of the present application;

[0020] Figure 2 is a schematic composition structure diagram of a disk array fault warning device provided by an embodiment of the present application;

[0021] Figure 3 is a schematic composition structure diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The terms in the specification, claims, and drawings of this application, such as "first", "second", "third", "fourth", etc. (if any), are used to distinguish similar objects and not to describe a specific order or sequence. Therefore, it is understood that, under appropriate circumstances, these terms can be used interchangeably so that the described embodiments can be implemented in different orders, unless there are special requirements in the drawings or descriptions. In addition, the terms "is" and "has" in this application and any of their variants are intended to non-exclusively cover all possible constituent elements. For example, a process, method, system, product, or device that includes several steps or units does not necessarily have to be limited to the steps or units explicitly listed, but may also include other steps or units not explicitly listed, or steps or units inherent to the process, method, product, or device.

[0023] In this application, a "module" or "unit" refers to a computer program or a part of a computer program with a specific function and works in cooperation with other relevant parts to achieve a predetermined goal. These modules or units can be implemented by software, hardware (such as a processing circuit or a memory), or a combination of both. One or more processors or memories can implement one or more modules or units. At the same time, each module or unit can also be a part of a larger module or unit.

[0024] The technical solutions in this application will be described in detail below with reference to the drawings in the embodiments. It should be noted that the described embodiments are only a part of this application, not all embodiments. In the following description, the "some embodiments" mentioned are only subsets of all possible embodiments, which can be the same or different subsets, and different embodiments can be combined with each other without conflict.

[0025] Figure 1 It is a schematic flowchart of a method for early warning of disk array failures provided by an embodiment of this application. Exemplarily, see Figure 1 The method for early warning of disk array failures provided by an embodiment of this application may include the following steps 101 to 104:

[0026] Step 101, in response to a preset periodic trigger condition, apply a preset mechanical excitation signal to the target disk array;

[0027] In some examples, the preset periodic trigger condition refers to the condition setting for triggering the detection process at regular intervals according to a predetermined time interval or operating state cycle. It can be set by the system administrator in the configuration file or control platform, or automatically adjusted according to at least one of the usage frequency, running duration, temperature, and load fluctuation of the disk array. For example, the preset periodic trigger condition can be a fixed cycle of triggering the detection every 8 hours, or an adaptive cycle of automatically triggering when the target disk array has accumulated 24 hours of operation or has been running continuously at a high temperature for more than 2 hours. The target disk array refers to the disk array system unit for which a fault warning operation needs to be performed currently, usually a set of hard disk collections with a logical structural relationship. The preset mechanical excitation signal refers to a physical vibration signal pre-generated according to a specific excitation scheme, used to stimulate the response of the disk array structure to vibration; it can be synthesized by the internal excitation signal module of the system according to a predetermined frequency range and amplitude, such as synthesizing a swept-frequency signal, a sine signal sequence, etc., for output to the vibrator on the disk array, such as a piezoelectric actuator.

[0028] Exemplarily, when the detection cycle arrives, the embedded control unit generates a composite swept-frequency excitation signal according to the set parameters, including a frequency band covering the operating frequency of the disk bearing and the resonance frequency of the head swing arm; the excitation signal is loaded onto the disk array bracket through the vibration driver, so that the target disk array can receive external mechanical disturbances on the premise of safety, thereby inducing an analyzable structural response signal.

[0029] By implementing step 101, an active detection mechanism can be achieved, avoiding passive waiting for the appearance of fault symptoms; through periodic excitation, detection can start when the operation is normal, improving the forward-looking and real-time nature of fault identification.

[0030] Step 102, collect the vibration response spectrum and acoustic response waveform of the target disk array under the action of the mechanical excitation signal;

[0031] In some examples, the vibration response spectrum refers to the frequency distribution map formed after the vibration signal generated by the target disk array under the action of a preset mechanical excitation signal undergoes frequency-domain transformation, which can be used to reflect the response characteristics of structural components in different frequency bands; the original vibration signal can be obtained through a piezoelectric acceleration sensor array arranged on the disk bracket or housing, and then signal processing algorithms (such as wavelet packet decomposition and spectrum analysis) are used to extract frequency components; for example, the appearance of a peak in the 10 kHz frequency band may indicate resonance anomalies of the head swing arm, and the appearance of a new resonance peak in the 3.5 kHz frequency band may suggest bearing wear problems. The acoustic response waveform refers to the acoustic wave signal generated by the target disk array under the action of a preset mechanical excitation signal, such as whistling and resonance sounds, which reflects the sound changes of the internal operating components of the device and can include time series and spectral characteristics; the sound signal can be collected through a multi-channel microphone array installed inside the target disk array or on the top of its housing, and then signal processing methods, such as Mel-Frequency Cepstral Coefficients (MFCC) transformation, are used to extract the characteristic waveform; for example, the generation of a continuous whistling sound near the disk rotation frequency may indicate eccentricity or rubbing problems during disk rotation; a short-term pulse that suddenly increases in the acoustic waveform may represent mechanical shock or looseness. The process of collecting the vibration response spectrum and the acoustic response waveform is to synchronously record the vibration and acoustic wave responses of the structure and housing of the target disk array to the excitation while applying the mechanical excitation signal, forming the basic data required for analysis; at the hardware level, real-time synchronous acquisition can be achieved through piezoelectric sensors and microphone arrays; at the software level, the sampling frequency, channel matching, and time synchronization can be controlled through the acquisition system, and the collected data can be preprocessed, filtered, and feature extracted.

[0032] Exemplarily, during the application of the mechanical excitation signal, the piezoelectric sensor array real-time collects the minute vibration responses at each physical disk bracket, and the sampling signal undergoes wavelet packet decomposition to form a multi-resolution spectrogram; at the same time, the multi-channel microphone array arranged on the top of the disk compartment synchronously collects the acoustic response signal, and the time series characteristics are extracted through the MFCC algorithm for subsequent fusion analysis and processing.

[0033] By implementing step 102, vibration and acoustic dual-channel signals are synchronously obtained, establishing a multi-modal response data foundation, which helps to capture physical anomalies in different dimensions; enhancing the recognition ability and sensitivity to different types of potential faults (such as bearing wear and swing arm resonance).

[0034] Step 103, fuse and analyze the vibration response spectrum and the acoustic response waveform through a preset feature fusion model to obtain a fault probability prediction vector;

[0035] In some examples, the preset feature fusion model is a pre-trained multi-modal deep learning model used to collaboratively process features from different sensing sources (such as vibration spectra and acoustic waveforms) to enhance the fault mode recognition ability. The preset feature fusion model can be constructed based on the Convolutional Neural Network (CNN) architecture, including multiple branch modules (such as vibration channels and acoustic channels), and can be trained using an existing fault data set in a supervised learning manner, with the optimization objective being the fault recognition accuracy or prediction precision. Fusion analysis refers to the process of uniformly expressing and correlating the data features of different modalities through the preset feature fusion model to comprehensively understand the system state, especially to capture abnormal manifestations that may be overlooked in a single modality. The fault probability prediction vector is a multi-dimensional vector output after the preset feature fusion model performs fusion analysis, and each dimension can represent the occurrence probability corresponding to a certain physical disk or fault type of the target disk array. The fault probability prediction vector can be generated from the feature tensor output by the fusion model through fully connected layers and activation functions such as softmax / sigmoid. The labels in the training stage come from manually annotated fault data sets or historical fault records.

[0036] Exemplarily, the preprocessed wavelet spectrogram and MFCC acoustic image can be respectively input into a dual-branch convolutional neural network to extract multi-layer features, and fused under the control of a dynamic attention mechanism to generate a fault representation tensor. Finally, the fault probability prediction vector output by the fully connected layer reflects the current state of the disk array, providing a quantitative basis for subsequent early warning decisions.

[0037] By implementing step 103, fusing multi-modal features improves the accuracy and robustness of fault recognition, and can overcome the problem that single-modal analysis is vulnerable to noise interference. Outputting the probability prediction vector helps to achieve quantitative fault assessment and supports subsequent hierarchical early warning and precise positioning.

[0038] Step 104, determine the real-time fault early warning instruction for the target disk array according to the fault probability prediction vector and the preset health benchmark vector;

[0039] In some examples, the preset health benchmark vector is a reference vector representing the multi-dimensional feature distribution in the "normal healthy state" of the disk array, which is used as a comparison object to determine whether the current state deviates from the normal state; the preset health benchmark vector can be generated through data accumulation of a disk array that has run without faults for a long period, or obtained by statistically averaging and normalizing a large number of "health samples" during the training stage of the preset feature fusion model, or a benchmark library can be established separately for disk arrays of different models, batches or operating environments. The real-time fault warning instruction is an operation command generated based on the deviation degree between the fault probability prediction vector and the preset health benchmark vector, which is used to indicate the triggering of corresponding warning levels or maintenance strategies; it can be determined by comparing the similarity between the fault probability prediction vector and the health benchmark vector, such as cosine similarity or Euclidean distance; according to the interval range where the similarity falls, warning instructions of different levels are generated, such as primary, intermediate, advanced, etc.; for example, when the cosine similarity > 0.95, it indicates that the state is healthy and no warning is required; when the cosine similarity drops below 0.70, a high-level warning is generated, prompting to immediately replace the disk and mark the physical location.

[0040] Exemplarily, the cosine similarity between the fault probability prediction vector and the corresponding health benchmark vector can be calculated, and whether to trigger a primary, intermediate or high-level warning can be automatically determined according to the threshold interval where the result falls; the generated warning instruction can be displayed through the console, pushed as an alarm and linked with the maintenance system to support intelligent response and visual operation and maintenance.

[0041] By implementing step 104, comparing the prediction result with the health benchmark can dynamically perceive the degree of deviation of the disk state; outputting clear warning instructions facilitates automatic linkage of the maintenance process (such as report generation, data migration, disk marking, etc.), improving the security and operation and maintenance efficiency of the target disk array.

[0042] In summary, through the fusion of the vibration response spectrum and the acoustic response waveform, the embodiment of the present application can obtain the state information of the disk array from multiple dimensions, overcome the problem of insufficient prediction accuracy of a single signal source, and improve the accuracy of fault identification and prediction; using an externally applied mechanical excitation instead of internal data monitoring does not affect the normal operation of the disk array, and has strong engineering adaptability and security; actively exciting and collecting response signals according to a preset period helps to identify potential anomalies earlier, improves the real-time performance and sensitivity of fault warning, and facilitates early intervention and maintenance; it can issue a warning before the disk array has a serious fault, helping the operation and maintenance personnel to handle problems in time, thereby reducing the economic losses caused by system downtime and improving the overall availability and reliability of the system. In summary, the disk array fault warning method provided by the embodiment of the present application can achieve high-precision, non-intrusive, real-time fault warning of the disk array through periodic mechanical excitation and multi-modal signal fusion analysis, improving the reliability of the disk array and reducing the operation and maintenance cost.

[0043] In some embodiments, applying the preset mechanical excitation signal to the target disk array may include: generating a composite swept-frequency signal having a first vibration frequency range and a second vibration frequency range, wherein the first vibration frequency range covers the rotational characteristic frequency band of the disk bearings of the target disk array, and the second vibration frequency range covers the resonant frequency band of the head swing arm of the target disk array; converting the composite swept-frequency signal into a preset mechanical excitation signal through an embedded controller; and time-sharingly loading the preset mechanical excitation signal to the vibration exciter of the target disk array based on a preset time series.

[0044] In some examples, the first vibration frequency range refers to the part of the composite swept-frequency signal that specifically covers the rotational characteristic frequency band of the disk bearing. The rotational characteristic frequency band of the disk bearing refers to the frequency range where mechanical vibration or noise is naturally generated by the internal bearing components of the disk drive during operation. The rotational characteristic frequency band of the disk bearing can be referenced by the calculation formulas for the characteristic frequencies of rolling bearings in the mechanical design manual (such as BPFO, BPFI), or determined by observing the response pattern of a long-term healthy device through spectrum measurement data. The first vibration frequency range is used to induce or amplify the response characteristics of bearing-like structures, and its typical operating frequency band can be calculated by analyzing the structural parameters of the disk bearing (such as ball diameter, rotational speed, load). For example, for a 7200 RPM (i.e., 120 Hz) hard disk, the bearing characteristic frequency range is generally in the interval of 200 Hz to 2 kHz. The second vibration frequency range is another frequency segment in the composite swept-frequency signal that specifically covers the resonant frequency band of the head swing arm. The resonant frequency band of the head swing arm is the frequency range where the head support (swing arm) exhibits a resonant response at a specific frequency, corresponding to its structural stiffness, mass, and boundary conditions, and is a key indicator for detecting head looseness, cracks, or fatigue failure. The second vibration frequency range can be used to stimulate the dynamic response of the head structure. As a flexible structure, the resonant frequency band of the head swing arm is usually above several kilohertz and can be obtained through finite element modeling, physical excitation tests, or experimental modal analysis. The composite swept-frequency signal is a continuous or segmented frequency signal with a wide frequency range, used to cover the response characteristics of multiple structural subsystems, and is composed of the first vibration frequency range and the second vibration frequency range. The composite swept-frequency signal can be generated using signal synthesizer software or hardware modules. For example, two linear swept-frequency signals: 200–2000 Hz + 3500–8000 Hz are synthesized separately and then merged. The generated composite swept-frequency signal can both stimulate the bearing response and excite the swing arm resonance. The preset time series refers to the control time plan for applying the excitation signal, used to control the loading time point, duration, and loading order of each frequency band of the excitation signal, which can be set through a configuration file, such as "load the first frequency band from 0 to 2 seconds, and load the second frequency band from 3 to 5 seconds", or a random time window can be used to improve the robustness of structural response recognition. The generated mechanical excitation signal can be loaded onto the vibrators (such as piezoelectric actuators) arranged at different positions of the disk array sequentially or alternately according to time segments to achieve local or global excitation of the structure. The excitation time and frequency band of each vibrator are controlled by an embedded control unit to implement the "time-domain distribution" loading strategy. For example, the bearing frequency band signal is alternately loaded onto the front-end exciter within 3 seconds, and then the swing arm frequency band signal is loaded onto the rear-end exciter.

[0045] Through the implementation of the above embodiments, composite swept-frequency excitation is performed on the frequency bands covering key structures such as disk bearings and head swing arms, making the excitation signal more targeted and comprehensive, capable of fully stimulating potential structural defects, improving the recognizability of the disk array fault warning response, and thus enhancing the sensitivity of fault prediction.

[0046] In some embodiments, the foregoing step 102 may include: collecting an original vibration signal of the target disk array under the action of a mechanical excitation signal through a piezoelectric sensor array disposed on a disk bracket of the target disk array; collecting an original acoustic signal of the target disk array under the action of the mechanical excitation signal through a multi-channel microphone array disposed on the top of a disk magazine of the target disk array; performing wavelet packet decomposition on the original vibration signal to obtain a vibration response spectrum; and performing Mel-frequency cepstral coefficient transformation on the original acoustic signal to obtain an acoustic response waveform including temporal features.

[0047] In some examples, the disk bracket is a structural support component inside the disk array for fixing physical hard disks, usually made of metal or composite materials, and has good mechanical coupling; the piezoelectric sensor array is a group of highly sensitive vibration detectors mounted on the disk bracket, which can convert minute mechanical strains into electrical signals, record mechanical responses, and an industrial-grade piezoelectric accelerometer array can be selected and arranged on different brackets to achieve multi-point collection; for example, two sensors are respectively arranged at the front end and the middle of each disk bracket to form an 8-channel array; the original vibration signal is an unfiltered or unprocessed voltage / current signal collected in real time by the piezoelectric sensor, reflecting the actual mechanical response of the structure under the excitation; the vibration response spectrum is a frequency-amplitude map obtained by performing wavelet packet decomposition (WPD) on the original vibration signal, revealing the energy distribution of the response signal in multiple frequency bands. The top of the disk magazine is the upper space area of the disk array housing, where an acoustic acquisition device can be arranged to facilitate the collection of sound signals propagated inside; a sensing window can be reserved or sound-absorbing materials can be installed on the top of the disk magazine to enhance the sound pickup effect; for example, three sound pickup holes are opened on the top of the disk array to install a multi-channel microphone array; the multi-channel microphone array is composed of multiple highly sensitive microphones, which can collect sound wave signals from multiple directions simultaneously; the original acoustic signal is unfiltered audio data collected by the microphone array, which can be a time-domain signal stored in PCM format, reflecting sound pressure changes; the acoustic response waveform is a two-dimensional temporal-spectrum image formed by performing Mel-frequency cepstral coefficient (MFCC) transformation on the original acoustic signal, including features related to auditory perception.

[0048] Exemplarily, while applying the mechanical excitation signal, the embedded system starts the sensor synchronous acquisition process: the piezoelectric array records the vibration of each bracket at a frequency of 25 kHz, and the microphone array records the audio signal inside the disk magazine at a frequency of 44.1 kHz; subsequently, three-layer wavelet packet decomposition is respectively performed on the vibration signal to extract the frequency band energy features; the MFCC transformation is applied to the acoustic signal to obtain a time-frequency joint feature map, providing multi-modal input for subsequent model analysis.

[0049] Through the implementation of the above embodiments, the piezoelectric sensor and the microphone array are combined to collect multi-dimensional original signals, and wavelet packet decomposition and mel cepstrum processing are used, which helps to extract more representative time-frequency and acoustic features and improve the resolution and stability of fault feature recognition.

[0050] In some embodiments, the foregoing preset feature fusion model may include: a first convolutional neural network branch for extracting time-frequency features from the vibration response spectrum and outputting a first feature tensor; a second convolutional neural network branch for extracting formant features from the acoustic response waveform and outputting a second feature tensor; and a dynamic attention fusion module for generating a fault probability feature vector according to the correlation weight of the first feature tensor and the second feature tensor.

[0051] In some examples, the first convolutional neural network branch is part of a preset feature fusion model, specifically designed to process data inputs from vibration response spectra and extract their time-frequency features, such as local energy changes, periodic patterns, resonance frequency bands, etc. The first convolutional neural network branch can be designed based on convolutional neural network structures such as Residual Network (ResNet) or Visual Geometry Group (VGG). The input is a wavelet packet spectrogram. The convolutional layer is used to extract spatially local features, followed by a pooling layer to compress information. For example, a vibration response spectrum with an input size of 64×64 outputs a first feature tensor with a size of 8×8×128 after 3 layers of convolution + pooling. The first feature tensor is an intermediate representation generated by the first convolutional neural network branch after processing the vibration response spectrum. It is a numerical tensor with a multi-dimensional structure and can include the number of channels (feature maps), time / frequency dimensions, etc. For example, a feature vector with a shape of [batch_size, 128] is output for subsequent fusion, where 128 represents the 128-dimensional deep features extracted from the vibration spectrum. The second convolutional neural network branch is another channel in the preset feature fusion model, dedicated to processing acoustic response waveforms and focusing on extracting auditory-related features such as formants and frequency shifts. The input is a Mel-frequency cepstral coefficient image, and it is modeled using a convolutional neural network structure similar to the first branch. For example, an acoustic response waveform image with an input dimension of 40×100 is extracted into a second feature tensor with a size of 16×25×64 through the convolutional network. The second feature tensor is a deep feature representation obtained by the second convolutional neural network branch after processing the acoustic response waveform and can contain resonance, impact, frequency structures, etc. under multiple acoustic modes. For example, after compression and flattening, it forms a 64-dimensional acoustic feature vector for easy fusion with vibration features. The dynamic attention fusion module is the central component connecting the outputs of the two convolutional neural network branches, used to model the semantic relationship between the first feature tensor and the second feature tensor, assign weights, and achieve effective fusion. The Transformer attention mechanism or SE-Block structure can be used to perform weighted fusion on the similarity between the two tensors. For example, by calculating the dot product similarity between the two features, normalizing it into attention weights, and finally forming a fused representation vector, and then outputting a vector containing probabilities of multiple fault categories through an activation function.

[0052] For example, the vibration response spectrum can be first input into the first CNN branch to extract multi-scale resonance features to obtain a 128-dimensional vibration feature vector; at the same time, the MFCC acoustic image is input into the second CNN branch to extract resonance peak features to form a 64-dimensional acoustic representation vector; then, the dynamic attention fusion module distributes the similarity weights between the two vectors and fuses them to generate a feature vector containing various potential fault probabilities as the basis for the fault warning system.

[0053] Through the implementation of the above embodiments, a dual-branch convolutional neural network is used to extract key features of different modalities, and weighted fusion is performed through a dynamic attention mechanism, which can achieve more effective feature association modeling and fault mode recognition, and greatly improve the prediction accuracy and robustness in complex fault scenarios.

[0054] In some embodiments, the aforementioned determination of the real-time fault warning instruction of the target disk array based on the failure probability prediction vector and the preset health reference vector may include: determining the cosine similarity between the failure probability feature vector and the preset health reference vector; when the cosine similarity is less than a first threshold and greater than or equal to a second threshold, the real-time fault warning instruction is to trigger a primary warning and generate a disk health report, wherein the second threshold is less than the first threshold; when the cosine similarity is less than the second threshold and greater than or equal to a third threshold, the real-time fault warning instruction is to trigger an intermediate warning and start redundant data migration, wherein the third threshold is less than the second threshold; when the cosine similarity is less than the third threshold, the real-time fault warning instruction is to trigger an advanced warning and mark the physical location of the faulty disk.

[0055] In some examples, the similarity between them can be measured by calculating the cosine value of the angle between the fault probability feature vector and the preset healthy reference vector, which is used to judge the deviation degree between the current state (fault probability feature vector) and the normal state (healthy reference vector). The first threshold, the second threshold, and the third threshold are similarity boundary parameters for hierarchical early warning, which are set in the system to trigger early warning responses at different levels. The setting condition is that the first threshold > the second threshold > the third threshold. These thresholds can be obtained from historical fault data and statistical learning methods, or set through expert experience. For example, the first threshold = 0.85, the second threshold = 0.70, and the third threshold = 0.50. A primary early warning indicates that the state has a slight anomaly, which may be a recoverable deviation. A secondary early warning indicates that there are obvious anomalies in the system and preventive measures are required. A high-level early warning indicates serious fault signs and immediate intervention or component replacement is required. Different levels of early warnings can be triggered by the interval into which the cosine similarity result falls. For example, if the similarity = 0.76, which is between the second and the first thresholds, a primary early warning is triggered. Under a primary early warning, no immediate intervention measures are taken, and only a disk health report is generated for operation and maintenance personnel to analyze and track trends. The disk health report may include: the current fault probability vector, the abnormal frequency position, the acoustic resonance feature map, etc., and a health score trend map is generated by comparing with historical reports. During a secondary early warning, data fault tolerance measures are initiated, and critical data is migrated from the disk that may have a fault to a redundant backup device or a hot spare disk. This action can call the data reconstruction process through a Redundant Array of Independent Disks (RAID) controller. For example, parity blocks are redistributed to the spare disk to improve data fault tolerance. A high-level early warning means that the disk is in a high-risk fault state. The physical location of the disk is immediately marked on the management interface or in the cabinet to prompt manual repair or replacement. The target disk can be made to blink the LED or the slot number can be displayed on the console interface through a control command, such as "Slot: B4, Status: Serious Fault".

[0056] Through the implementation of the above embodiments, different cosine similarity thresholds are set to implement a three-level early warning mechanism, which not only improves the refinement level of early warning responses but also automatically triggers corresponding processing measures according to different risk levels, such as health report generation, data migration, disk marking, etc., enhancing the initiative and intelligence of fault response.

[0057] In some embodiments, it may further include: generating a physical location code for each physical disk in the target disk array according to the preset physical topology structure of the target disk array; determining the fault probability value of the physical disk according to the sub-vector component corresponding to each physical disk in the target disk array in the fault probability prediction vector; and determining the historical fault similarity of each physical disk in the target disk array according to the matching result between the vibration response spectrum and the historical fault case library.

[0058] In some examples, the preset physical topology refers to the spatial layout and logical mapping relationship of physical disks in the target disk array, describing the actual installation position, hierarchical number, and channel connection method of each physical disk in the overall array; the preset physical topology can be obtained through disk array hardware configuration files, redundant array of independent disks (RAID) controllers, cabinet layout tables, etc. For example, the topology of a certain array is a 4×3 layout, with slot numbers ranging from A1 to C4, and the corresponding controller and channel information such as "Ctrl0-Ch1-SlotB3"; the physical location encoding of each physical disk is a unique encoding used to identify the specific position of the disk in the array, usually generated by combining the row and column of the cabinet, slot number, channel number, etc. Each component in the failure probability prediction vector can correspond to a physical disk, representing its individual failure probability, and the sub-vector component represents the risk value of the specific disk; for example, in an array with 12 disks, the corresponding vector is [0.02, 0.04, 0.85, 0.07, …, 0.01], where the third component 0.85 indicates that the failure probability of the disk in slot "B3" is 85%. The failure probability value of a physical disk is the probability result determined according to the sub-vector component corresponding to each physical disk in the target disk array in the failure probability prediction vector, representing the risk intensity of the current disk failure, with a range of 0 to 1. By comparing the currently collected disk vibration spectrum features with the historically labeled failure spectrum samples, potential similar failure modes can be identified. The historical failure similarity of each physical disk in the target disk array represents the similarity between the current disk vibration characteristics and a specific historical failure (such as head impact, bearing defect), which can be used to assist in the preliminary judgment of the failure type.

[0059] Exemplarily, after the failure prediction model outputs a vector, the physical disk position encoding corresponding to each component can be automatically parsed according to the topology of the disk array, and the failure probability value of each disk can be generated; subsequently, the current vibration spectrum features are extracted and matched one by one with various known failure samples in the historical failure case library to calculate the similarity index; by combining the probability value and the similarity result, not only can high-risk disks be marked, but also their possible failure types can be preliminarily inferred for subsequent decision-making reference and the generation of maintenance suggestions.

[0060] Through the implementation of the above embodiments, by combining physical topology encoding, sub-vector failure probability, and historical failure matching, the status of each physical disk in the disk array can be independently evaluated and located, enhancing the interpretability and operability of early warning, and providing a basis for subsequent precise maintenance and replacement.

[0061] In some embodiments, it may further include: displaying a three-dimensional fault hot spot distribution map in a graphical user interface, where the X-axis of the three-dimensional fault hot spot distribution map represents the physical location code, the Y-axis represents the fault probability value, and the Z-axis represents the historical fault similarity of each physical disk in the target disk array; determining the mapping color of the three-dimensional fault hot spot distribution map according to the warning level in the real-time fault warning instruction.

[0062] In some examples, the graphical user interface is a visual interface for operation and maintenance personnel, engineers, etc. to interact with the system, and may include functions such as data display, operation control, and warning feedback; the graphical user interface can be built by front-end development tools (such as Qt, Electron, Vue.js, etc.). For example, the graphical user interface of the disk array health monitoring system includes: a fault probability map, a disk status panel, a warning log list, etc. The three-dimensional fault hot spot distribution map is a three-dimensional visualization graph used to intuitively display the health status, fault risk, and historical feature matching of each disk in the disk array in three dimensions, and can be drawn using a 3D graphics library. For example, the display form is a three-dimensional bar chart or a scatter plot, and the user can rotate and zoom the perspective to view the disk array fault distribution. The X-axis is used to mark the physical location of the disk in the array (such as "SlotA1", "B3", "C4", etc.), that is, the location code, and the physical layout information can be read from the array controller or the configuration file, for example, sorted by row and column numbers or slot numbers; the Y-axis represents the current fault probability value predicted by the fusion model, and the range is generally from 0 to 1, which can be directly extracted from the fault probability prediction vector output by the model. For example, the probability of the "SlotB3" disk is 0.87; the Z-axis represents the similarity score between the current vibration or acoustic characteristics and the historical fault samples, which is used to assist in judging the possible type of fault, and can be obtained through cosine similarity, Euclidean distance, or a deep feature comparison model. For example, the similarity score of "SlotA2" is 0.91, which highly matches the "bearing wear" sample. Color mapping can be combined with the level output by the warning instruction, and different colors represent different warning levels, which is used to enhance the intuitive expression of the graph and the risk identification ability. For example, green represents normal (similarity > 0.9 and fault probability < 0.2), yellow represents a primary warning, orange represents an intermediate warning, and red represents a high-level warning.

[0063] Through the implementation of the above embodiments, using three-dimensional graphical display of physical location, fault probability, and historical similarity, the fault hot spots can be made visible, controllable, and traceable. Combining the color mapping of different warning levels enables maintenance personnel to quickly identify high-risk disks and take measures, significantly improving the operation and maintenance efficiency and experience.

[0064] In some embodiments, it may further include: dynamically generating a fault evolution trend graph of the target disk array based on the multi-dimensional data in the three-dimensional fault hot spot distribution graph and combining time series analysis, for evaluating the fault development rate and risk accumulation situation.

[0065] In some examples, the fault evolution trend graph is a line graph or a heat map with time as the horizontal axis and the change in fault probability or similarity as the vertical axis, for showing the state fluctuations of each physical disk within a certain time window; the data of the past 24 hours, 72 hours or the recent week can be aggregated and analyzed through a sliding window, and it supports users to select specific slot disks for retrospective viewing; for example, the fault probability of a certain disk has rapidly increased from 0.15 to 0.82 within the past 48 hours, and at the same time the similarity has increased from 0.40 to 0.89, indicating that its state is rapidly evolving towards a typical fault mode.

[0066] Exemplarily, after the system detects a primary or intermediate warning for a certain disk, the background continuously records the change in the fault probability and similarity of the disk, and refreshes the trend graph data every 5 minutes; the graphical user interface displays the probability increase curve of the disk within the past several hours through a curve graph, and overlays a color interval background to assist in understanding different risk levels; in addition, the change trend of the vibration energy spectrum can also be overlaid to provide a more comprehensive basis for trend judgment.

[0067] Through the implementation of the above embodiments, the system not only provides the current static risk judgment, but also reveals the fault evolution path and development speed through trend visualization means, enabling the operation and maintenance personnel to take preventive maintenance measures before the problem deteriorates and improving the forward-looking perception ability of potential faults.

[0068] In some embodiments, a continuous learning mechanism can be introduced into the fault warning system to adapt to the long-term changes in the operating environment of the disk array and the emergence of new fault modes.

[0069] In some examples, the continuous learning mechanism includes: dynamically annotating and storing new data samples continuously accumulated during actual operation, and periodically performing incremental training on the fault recognition model; the annotation method can combine means such as expert review, operation log association or model prediction result feedback optimization to achieve a semi-automated label update process; the model update can be based on strategies such as transfer learning, incremental learning or federated learning to avoid forgetting existing knowledge while integrating new knowledge.

[0070] Exemplarily, the system automatically extracts disk samples that triggered warnings within the past month at regular intervals, combines subsequent maintenance records or expert verification results for label confirmation, and adds them to the extended training set; subsequently, a fine-tuning strategy is used to train for several rounds on the basis of the original model to make the model parameters adapt to the new data feature distribution and further improve the recognition ability for the latest fault types.

[0071] Through the implementation of the above embodiments, the system can dynamically adapt to changes in the operating environment of the disk array, avoid model degradation, improve the response ability and generalization ability of the fault detection model to rare anomalies and new types of faults, and continuously optimize the early warning accuracy and practicality.

[0072] Furthermore, as an implementation of the foregoing method embodiments, the present application also provides a disk array fault early warning device for implementing the foregoing method embodiments. The device embodiments correspond to the foregoing method embodiments. For the convenience of reading, the details of the foregoing method embodiments will not be repeated one by one in the disk array fault early warning device embodiments, but it should be clear that the devices in the embodiments of the present application can correspondingly implement all the contents of the foregoing method embodiments. As Figure 2 shown, the disk array fault early warning device 20 includes: a signal application unit 201, a signal acquisition unit 202, a vector prediction unit 203, and an instruction determination unit 204. Among them, the signal application unit 201 is configured to apply a preset mechanical excitation signal to the target disk array in response to a preset periodic trigger condition; the signal acquisition unit 202 is configured to acquire the vibration response spectrum and the acoustic response waveform of the target disk array under the action of the mechanical excitation signal; the vector prediction unit 203 is configured to perform fusion analysis on the vibration response spectrum and the acoustic response waveform through a preset feature fusion model to obtain a fault probability prediction vector; the instruction determination unit 204 is configured to determine a real-time fault early warning instruction for the target disk array according to the fault probability prediction vector and a preset health reference vector.

[0073] In some embodiments, the signal application unit 201 is further configured to generate a composite swept-frequency signal having a first vibration frequency range and a second vibration frequency range, where the first vibration frequency range covers the disk bearing rotation characteristic frequency band of the target disk array, and the second vibration frequency range covers the head swing arm resonance frequency band of the target disk array; convert the composite swept-frequency signal into a preset mechanical excitation signal through an embedded controller; and time-divisionally load the preset mechanical excitation signal onto the vibration exciter of the target disk array based on a preset time series.

[0074] In some embodiments, the signal acquisition unit 202 is further configured to acquire the original vibration signal of the target disk array under the action of the mechanical excitation signal through a piezoelectric sensor array disposed on the disk bracket of the target disk array; acquire the original acoustic signal of the target disk array under the action of the mechanical excitation signal through a multi-channel microphone array disposed on the top of the disk compartment of the target disk array; perform wavelet packet decomposition on the original vibration signal to obtain a vibration response spectrum; and perform Mel cepstral coefficient transformation on the original acoustic signal to obtain an acoustic response waveform including time series features.

[0075] In some embodiments, the preset feature fusion model includes: a first convolutional neural network branch for performing time-frequency feature extraction on the vibration response spectrum and outputting a first feature tensor; a second convolutional neural network branch for performing formant feature extraction on the acoustic response waveform and outputting a second feature tensor; and a dynamic attention fusion module for generating a fault probability feature vector according to the correlation weight between the first feature tensor and the second feature tensor.

[0076] In some embodiments, the instruction determination unit 204 is further configured to determine the cosine similarity between the fault probability feature vector and a preset health reference vector; when the cosine similarity is less than the first threshold and greater than or equal to the second threshold, the real-time fault warning instruction is to trigger a primary warning and generate a disk health report, where the second threshold is less than the first threshold; when the cosine similarity is less than the second threshold and greater than or equal to the third threshold, the real-time fault warning instruction is to trigger an intermediate warning and initiate redundant data migration, where the third threshold is less than the second threshold; when the cosine similarity is less than the third threshold, the real-time fault warning instruction is to trigger a high-level warning and mark the physical location of the faulty disk.

[0077] In some embodiments, the disk array fault warning device 20 further includes a user display unit, configured to generate physical location codes for each physical disk in the target disk array according to the preset physical topology of the target disk array; determine the fault probability value of the physical disk according to the sub-vector component corresponding to each physical disk in the target disk array in the fault probability prediction vector; and determine the historical fault similarity of each physical disk in the target disk array according to the matching result between the vibration response spectrum and the historical fault case library.

[0078] In some embodiments, the user display unit is further configured to display a three-dimensional fault hot spot distribution map in the graphical user interface, where the X-axis of the three-dimensional fault hot spot distribution map represents the physical location code, the Y-axis represents the fault probability value, and the Z-axis represents the historical fault similarity of each physical disk in the target disk array; and determine the mapping color of the three-dimensional fault hot spot distribution map according to the warning level in the real-time fault warning instruction.

[0079] The present application also provides a computer-readable storage medium storing computer-executable instructions or a computer program, which when executed by a processor, will cause the processor to execute any step of the disk array fault warning method provided by the present application.

[0080] In some embodiments, the computer-readable storage medium may be a memory such as a random access memory (RAM), a read-only memory (ROM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); or it may be various devices including one or any combination of the above memories.

[0081] In some embodiments, the computer-executable instructions may be in the form of a program, software, a software module, a script, or code, written in any form of programming language (including a compiled or interpreted language, or a declarative or procedural language), and may be deployed in any form, including being deployed as an independent program or being deployed as a module, a component, a subroutine, or other units suitable for use in a computing environment.

[0082] In some embodiments, the computer-executable instructions may or may not correspond to a file in the file system, and may be stored as part of a file that stores other programs or data. For example, they may be stored in one or more scripts in a hypertext markup language (HTML) document, stored in a single file dedicated to the program in question, or stored in multiple cooperating files (such as files that store one or more modules, subroutines, or code portions).

[0083] In some embodiments, the computer-executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one location, or, on multiple electronic devices distributed at multiple locations and interconnected by a communication network.

[0084] As Figure 3 shown, the present application further provides an electronic device 30, including a memory 310, a processor 320, and a computer program 311 stored on the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, any step of the above disk array fault warning method is implemented.

[0085] The present application further provides a computer program product, which includes a computer program or computer-executable instructions. The computer program or computer-executable instructions are stored in a computer-readable storage medium. The processor of the electronic device reads the computer program or computer-executable instructions from the computer-readable storage medium, and the processor executes the computer program or computer-executable instructions, so that the electronic device executes any step of the disk array fault warning method of the present application.

[0086] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting disk array failures, characterized in that Including: In response to a preset periodic triggering condition, a composite swept-frequency signal with a first vibration frequency range and a second vibration frequency range is generated, where the first vibration frequency range covers the disk bearing rotation characteristic frequency band of the target disk array, and the second vibration frequency range covers the head swing arm resonance frequency band of the target disk array; The composite swept-frequency signal is converted into a preset mechanical excitation signal through an embedded controller; Based on a preset time series, the preset mechanical excitation signal is loaded into the vibration exciter of the target disk array in a time-sharing manner; Collect the vibration response spectrum and acoustic response waveform of the target disk array under the action of the mechanical excitation signal; Perform fusion analysis on the vibration response spectrum and acoustic response waveform through a preset feature fusion model to obtain a fault probability prediction vector; According to the fault probability prediction vector and a preset health benchmark vector, determine a real-time fault warning instruction for the target disk array.

2. The disk array fault warning method according to claim 1, characterized in that, The collecting the vibration response spectrum and acoustic response waveform of the target disk array under the action of the mechanical excitation signal includes: Collect the original vibration signal of the target disk array under the action of the mechanical excitation signal through a piezoelectric sensor array arranged on the disk bracket of the target disk array; Collect the original acoustic signal of the target disk array under the action of the mechanical excitation signal through a multi-channel microphone array arranged on the top of the disk compartment of the target disk array; Perform wavelet packet decomposition on the original vibration signal to obtain the vibration response spectrum; Perform Mel cepstral coefficient transformation on the original acoustic signal to obtain the acoustic response waveform including time series characteristics.

3. The disk array fault warning method according to claim 1, wherein, The preset feature fusion model includes: A first convolutional neural network branch for extracting time-frequency features from the vibration response spectrum and outputting a first feature tensor; A second convolutional neural network branch for extracting formant features from the acoustic response waveform and outputting a second feature tensor; A dynamic attention fusion module for generating the fault probability prediction vector according to the correlation weight between the first feature tensor and the second feature tensor.

4. The disk array fault warning method according to claim 1, characterized in that, The determining the real-time fault warning instruction for the target disk array according to the fault probability prediction vector and a preset health benchmark vector includes: Determine the cosine similarity between the fault probability prediction vector and the preset health benchmark vector; When the cosine similarity is less than the first threshold and greater than or equal to the second threshold, the real-time fault warning instruction is to trigger a primary warning and generate a disk health report, where the second threshold is less than the first threshold; When the cosine similarity is less than the second threshold and greater than or equal to the third threshold, the real-time fault warning instruction is to trigger an intermediate warning and start redundant data migration, where the third threshold is less than the second threshold; When the cosine similarity is less than the third threshold, the real-time fault warning instruction is to trigger a high-level warning and mark the physical location of the faulty disk.

5. The disk array fault warning method according to claim 1, wherein It also includes: Generate physical location codes for each physical disk in the target disk array according to the preset physical topology structure of the target disk array; Determine the failure probability value of the physical disk according to the sub-vector components corresponding to each physical disk in the target disk array in the failure probability prediction vector; Determine the historical failure similarity of each physical disk in the target disk array according to the matching result between the vibration response spectrum and the historical failure case library.

6. The disk array fault warning method according to claim 5, wherein Further includes: Display a three-dimensional failure hot spot distribution map in the graphical user interface, where the X-axis of the three-dimensional failure hot spot distribution map represents the physical location code, the Y-axis represents the failure probability value, and the Z-axis represents the historical failure similarity of each physical disk in the target disk array; Determine the mapping color of the three-dimensional failure hot spot distribution map according to the warning level in the real-time failure warning instruction.

7. A disk array fault warning device, characterized in that, Includes: A signal application unit, configured to generate a composite sweep signal with a first vibration frequency range and a second vibration frequency range in response to a preset periodic trigger condition, where the first vibration frequency range covers the disk bearing rotation characteristic frequency band of the target disk array, and the second vibration frequency range covers the head swing arm resonance frequency band of the target disk array; convert the composite sweep signal into a preset mechanical excitation signal through an embedded controller; and time-division load the preset mechanical excitation signal onto the vibration exciter of the target disk array based on a preset time series; A signal acquisition unit, configured to acquire the vibration response spectrum and the acoustic response waveform of the target disk array under the action of the mechanical excitation signal; A vector prediction unit, configured to perform fusion analysis on the vibration response spectrum and the acoustic response waveform through a preset feature fusion model to obtain a failure probability prediction vector; An instruction determination unit, configured to determine a real-time failure warning instruction for the target disk array according to the failure probability prediction vector and a preset health benchmark vector.

8. An electronic device, comprising: A memory and a processor, characterized in that when the processor executes the computer program stored in the memory, the steps of the disk array failure warning method according to any one of claims 1-6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the disk array failure warning method according to any one of claims 1-6 are implemented.

Images