A high-frequency vibration signal cooperative processing method and system

Abstract

The application belongs to the technical field of power systems, and discloses a high-frequency vibration signal cooperative processing method and system, which comprises the following steps: receiving a high-frequency vibration signal and writing the signal into a real-time buffer according to a signal channel to generate waveform statistical features; performing dynamic subscription and dynamic unsubscription operations on the signal channel, applying a delay operation to the dynamic unsubscription operation, reading target waveform data in a time window, and generating a set of intercepted segments; performing batch learning and coordinated processing on the set of intercepted segments according to an optimal path to obtain batch learning samples; writing the waveform statistical features, the set of intercepted segments and the batch learning samples into an asynchronous archiving queue and generating an archiving file; in an offline stage, performing intersection interception operation on a target interception interval and an original data block interval determined by a data block index in the archiving file to generate an offline segment file for offline review, export and reanalysis. The application reduces invalid event processing and full-amount rule evaluation, and improves system response efficiency.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a method and system for collaborative processing of high-frequency vibration signals. Background Technology

[0002] With the development of modern industry and the advancement of technology, high-frequency vibration signals are playing an increasingly important role in equipment condition monitoring in fields such as machinery manufacturing, aerospace, and energy development. High-frequency vibration signals carry crucial information about the operating status of mechanical equipment, providing core data support for fault diagnosis, performance evaluation, and predictive maintenance.

[0003] Existing solutions for processing high-frequency vibration signals from industrial equipment generally suffer from problems such as link resource contention, unstable triggering logic, low data processing efficiency, and insufficient offline interception accuracy. These shortcomings are particularly pronounced when deployed in cloud-edge collaborative architectures or industrial internet platforms. 1) The online display and analysis adopts the full data reading method, which involves a large amount of data transfer. This not only makes it easy to compete with the front-end acquisition link for CPU, memory and bus resources, resulting in a decrease in real-time performance and link blockage at the edge, but also causes redundant data transmission between the cloud and the edge, occupying the bandwidth of the industrial internet platform.

[0004] 2) Conditional triggering links rely on fixed thresholds and instantaneous state judgments. Under unstable operating conditions such as short-term equipment jitter and load fluctuations, they are prone to false starts and stops, resulting in poor trigger reliability. If such unreliable triggering fragments are uploaded to the cloud, it will further pollute the cloud sample library and reduce the overall analysis quality of the platform.

[0005] 3) In multi-threshold, multi-task learning scenarios, each task independently loads historical samples, resulting in frequent repetitive I / O operations, high storage and computing overhead, and low overall system efficiency. In the cloud-edge collaborative mode, the edge needs to respond to local real-time tasks and training tasks sent from the cloud at the same time, which amplifies the problem of repeated reading and leads to waste of edge computing power and storage resources.

[0006] 4) Offline secondary interception lacks standardized and high-precision interval operation logic, which is prone to data misalignment and timestamp mismatch, destroying the original data time semantics and affecting the accuracy of fault diagnosis and review. When the data is synchronized and archived multiple times by the cloud and edge, the problem of time semantic disorder becomes more prominent, making it difficult for the industrial Internet platform to achieve unified data governance across nodes.

[0007] The above-mentioned problems are interconnected. Existing technologies have not formed an integrated collaborative mechanism that connects online real-time processing, intelligent learning and archiving, and accurate offline verification, making it difficult to meet the processing requirements of high-end equipment condition monitoring for high-frequency vibration signals.

[0008] Therefore, how to provide a method and system for collaborative processing of high-frequency vibration signals is an urgent problem to be solved. Summary of the Invention

[0009] This invention provides a method and system for collaborative processing of high-frequency vibration signals to solve the problems mentioned above in the prior art.

[0010] According to a first aspect of the present invention, a method for collaborative processing of high-frequency vibration signals is provided.

[0011] In one embodiment, the high-frequency vibration signal collaborative processing method includes: The system receives high-frequency vibration signals and writes them into a real-time buffer according to the signal channel. It then performs waveform statistical calculations on the real-time buffer to generate waveform statistical features. Based on the waveform statistical characteristics, the signal channel is dynamically subscribed and unsubscribed, and a delay operation is applied to the dynamic unsubscription operation to determine the interception time window, read the target waveform data within the interception time window, and generate an interception segment set. Based on the set of requests to be processed, the truncated fragment set is processed in batch learning coordination according to the optimal path to obtain batch learning samples. The waveform statistical features, the truncated fragment set, and the batch learning samples are written into an asynchronous archiving queue and an archive file is generated; wherein, the data block index in the archive file determines the original data block range; During the offline phase, an intersection truncation operation is performed on the target truncation range and the original data block range determined by the data block index in the archive file to generate an offline fragment file for offline review, export, and reanalysis.

[0012] In one embodiment, receiving high-frequency vibration signals and writing them into a real-time buffer according to signal channels, and performing waveform statistical calculations on the real-time buffer to generate waveform statistical features includes: High-frequency vibration signals are acquired, the high-frequency vibration signals are divided into partitions according to signal channels, and the partitioned signal data is stored in a real-time buffer to form channel data blocks. By using single instruction multiple data to perform waveform statistical calculations on each channel data block in the real-time buffer in parallel, the mean, variance and waveform factor of each channel data block are calculated within the sliding time window to generate a basic statistical feature set. Based on the basic statistical feature set, the amplitude characteristics of each signal channel within the current sliding time window are calculated, and the corresponding feature timestamps are marked. The amplitude features are correlated with the feature timestamps in a time series to generate structured waveform statistical feature data.

[0013] In one embodiment, the step of dynamically subscribing to and unsubscribing from the signal channel based on waveform statistical characteristics, applying a delay to the dynamic unsubscribing operation, determining a truncation time window, reading the target waveform data within the truncation time window, and generating a set of truncation segments includes: The waveform statistical feature data is compared with the preset trigger conditions in real time, and the signal channel is dynamically subscribed and unsubscribed based on the comparison results. The dynamic unsubscribe operation is delayed and verified through a delayed scheduling mechanism. Based on the timestamp of the dynamic subscription and the timestamp of the dynamic unsubscribe after delay verification, the start and end boundaries of the intercept time window are determined, and the intercept time window is generated. Read the target waveform data within the capture time window, and reconstruct the target waveform data according to the signal channel dimension to generate a capture segment set.

[0014] In one embodiment, the step of comparing waveform statistical feature data with preset trigger conditions in real time, and performing dynamic subscription and unsubscription operations on the signal channel based on the comparison results, includes: Set the signal channel to a silent standby state and monitor in real time whether the waveform statistical characteristic data meets the preset trigger conditions; If the preset triggering conditions are met, a dynamic subscription operation is performed on the signal channel, a dynamic subscription event is generated, and the timestamp of the dynamic subscription event is recorded to determine the starting boundary of the interception time window. Based on dynamic subscription events, the operating status of subscribed signal channels is continuously monitored, and waveform statistical feature data is compared with preset unsubscription thresholds in real time to determine whether the operating conditions of the equipment have fallen back to the preset range. If the operating conditions do not fall back to the preset range, the signal channel will be dynamically subscribed again, and the operating status of the subscribed signal channel will be continuously monitored. If the operating conditions fall back to the preset range, the signal channel will be dynamically unsubscribed.

[0015] In one embodiment, the step of delaying and verifying the dynamic unsubscribe operation through a delayed scheduling mechanism, and determining the start and end boundaries of the truncation time window based on the timestamp of the dynamic subscription and the timestamp of the delayed-verified dynamic unsubscribe, and generating the truncation time window includes: Set a delay waiting window for the signal channel that triggers the dynamic unsubscribe operation; The waveform statistical characteristics within the delay waiting window are verified, and the unsubscribe verification result is generated. If the unsubscribe verification result is a dynamic unsubscribe event, then extract the timestamp corresponding to the dynamic unsubscribe event and extract the termination boundary of the time window. If the unsubscribe verification result is not a dynamic unsubscribe event, then cancel the delayed unsubscribe operation, maintain the subscription status of the signal channel, and continue monitoring; The time period is defined by the candidate values ​​of the start and end boundaries, and a time window is generated.

[0016] In one embodiment, verifying the waveform statistical characteristics within the delay waiting window and generating an unsubscribe verification result includes: The original waveform data segments are extracted from the delay waiting window and segmented according to the preset time granularity based on the sliding window mechanism to generate a waveform segment dataset. Multi-dimensional statistical feature extraction processing is performed on the waveform segment dataset. The multi-dimensional statistical feature extraction processing includes using the fast Fourier transform algorithm to convert the time domain signal in the waveform segment dataset into the frequency domain signal, extracting the main frequency energy ratio, spectral entropy and frequency band energy distribution features, and generating a multi-dimensional statistical feature vector set. The multi-dimensional statistical feature vector set is compared with the dynamic threshold benchmark to generate a deviation assessment result; The stability score of the deviation evaluation results is calculated using a weighted fusion algorithm, and the unsubscription verification result is generated.

[0017] In one embodiment, the step of performing batch learning coordination processing on the truncated fragment set according to the optimal path based on the set of requests to be processed, to obtain batch learning samples, includes: Receive several sample interval requests corresponding to each learning task, wherein the sample interval request includes a target time range and a signal channel identifier, and generate a set of requests to be processed. The system summarizes, deduplicates, and merges all sample intervals in the request set to be processed, merging overlapping or adjacent sample intervals into continuous sample intervals to generate a list of read requests. Based on the time order and channel distribution in the read demand list, combined with the physical layout of the storage medium, the optimal read path is calculated and the optimal access plan is generated. According to the optimal access plan, the required sample data is read from the truncated fragments, and the reading results are uniformly saved in memory as a shared sample set to generate batch learning samples.

[0018] In one embodiment, the step of calculating the optimal read path and generating the optimal access plan based on the time order and channel distribution in the read demand list, combined with the physical layout of the storage medium, includes: For each read task in the read request list, schedule and sort them according to task priority, time urgency or data size to generate a priority-scheduled read sequence; Based on the priority-scheduled read sequence, read operations of adjacent sample intervals or the same storage channel are merged to generate a merged batch input / output request list; By combining the physical sector layout of the storage medium and the cache hit status, the optimal reading order of the batch input / output request list is calculated, and the optimal access plan is generated.

[0019] In one embodiment, the offline stage, performing an intersection truncation operation on the target truncation interval and the original data block interval determined by the data block index in the archive file to generate an offline fragment file for offline review, export, and reanalysis, includes: Receive the target cut-off range input by the user, and retrieve the original data block range and data block index that overlap with the target cut-off range in the archive file to generate the block original dataset to be processed; Determine whether each original data block in the original dataset to be processed overlaps with the target time range, filter out non-overlapping data blocks, extract valid data segments with overlap, and generate a list of filtered valid data segments. The data segments in the filtered list of valid data segments are sorted and concatenated in chronological order, and timestamps are added at the junctions of adjacent data segments to generate calibrated continuous time-series data. The calibrated continuous time series data is packaged into a standard offline file format to generate offline fragment files for offline verification, export, and reanalysis.

[0020] According to a second aspect of the present invention, a high-frequency vibration signal collaborative processing system is provided.

[0021] In one embodiment, a high-frequency vibration signal collaborative processing system includes: The signal receiving module is used to receive high-frequency vibration signals and write them into the real-time buffer according to the signal channel, and to perform waveform statistical calculations on the real-time buffer to generate waveform statistical features. The waveform calculation module is used to dynamically subscribe to and unsubscribe from the signal channel based on the waveform statistical characteristics, apply a delay to the dynamic unsubscribe operation, determine the truncation time window, read the target waveform data within the truncation time window, and generate a set of truncation segments. The learning coordination module is used to perform batch learning coordination processing on the truncated fragment set according to the optimal path based on the set of requests to be processed, so as to obtain batch learning samples. The feature summarization module is used to write waveform statistical features, truncated fragment sets, and batch learning samples into an asynchronous archiving queue and generate an archive file; wherein the data block index in the archive file determines the original data block range; The offline management module is used to perform intersection and truncation operations on the target truncation range and the original data block range determined by the data block index in the archive file during the offline stage, generating offline fragment files for offline review, export, and re-analysis.

[0022] According to a third aspect of the present invention, a computer device is provided.

[0023] In some embodiments, the computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described above.

[0024] According to a fourth aspect of the present invention, a computer-readable storage medium is provided.

[0025] In one embodiment, a computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the steps of the above method.

[0026] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: 1) This invention uses incremental reading based on timestamp boundaries and binary positioning to reduce the complexity of data retrieval at the edge, thereby reducing the overhead of data reading and transportation, and supporting the efficient synchronization of incremental data from edge nodes to the cloud in cloud-edge collaboration.

[0027] 2) This invention utilizes SIMD parallel instructions to accelerate waveform statistical calculations, improving the statistical throughput per unit time in scenarios with large samples and high sampling rates, and providing real-time and reliable vibration characteristic data for industrial internet platforms.

[0028] 3) This invention reduces invalid event processing and full rule evaluation by using dynamic subscription in conjunction with inverted index mapping, thereby improving system response efficiency; it also effectively filters short-term jitter through a cancelable delay scheduling mechanism, solving the problem of false start and false stop from the root, ensuring the quality of the truncated segments triggered at the edge, and providing reliable samples for cloud-based intelligent analysis.

[0029] 4) This invention achieves sample sharing and loading through batch learning coordination, eliminating redundant I / O and redundant calculations, and improving the efficiency of multi-task learning.

[0030] 5) This invention reduces the risk of real-time link blocking by asynchronous archiving and byte-mapped disk writing, improves system stability and data throughput, and achieves efficient persistence of data to the cloud or data lake.

[0031] 6) This invention ensures the temporal consistency of offline reconstructed data by intersecting intervals, fully preserving the original temporal semantics, supporting offline verification and cross-node reanalysis in the cloud, and improving the accuracy of data governance and fault diagnosis of industrial internet platforms.

[0032] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0033] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0034] Figure 1 This is a flowchart illustrating a high-frequency vibration signal collaborative processing method according to an exemplary embodiment; Figure 2 This is a schematic diagram of a high-frequency vibration signal collaborative processing system according to an exemplary embodiment; Figure 3 This is a schematic diagram of the structure of a computer device according to an exemplary embodiment; Figure 4 This is a flowchart illustrating the dynamic subscription state switching process according to an exemplary embodiment; Figure 5 This is a delayed start / stop timing diagram illustrated according to an exemplary embodiment; Figure 6 This is a flowchart illustrating a batch learning coordination process according to an exemplary embodiment. Detailed Implementation

[0035] The following description and accompanying drawings fully illustrate specific embodiments described herein to enable those skilled in the art to practice them. Some portions and features of certain embodiments may be included in or replace portions and features of other embodiments. The scope of the embodiments herein includes the entire scope of the claims and all available equivalents thereof. The various embodiments described herein are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0036] The modules in the apparatus or system of this application can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0037] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0038] Figure 1 An embodiment of a high-frequency vibration signal collaborative processing method of the present invention is shown.

[0039] In this optional embodiment, the high-frequency vibration signal collaborative processing method includes: Step S101: Receive high-frequency vibration signals and write them into the real-time buffer according to the signal channel; perform waveform statistical calculations on the real-time buffer to generate waveform statistical features. In this optional embodiment, the process of receiving high-frequency vibration signals and writing them into a real-time buffer according to signal channels, and performing waveform statistical calculations on the real-time buffer to generate waveform statistical features includes: High-frequency vibration signals are acquired, the high-frequency vibration signals are divided into partitions according to signal channels, and the partitioned signal data is stored in a real-time buffer to form channel data blocks. The waveform statistical calculations of each channel data block in the real-time buffer are performed in parallel using the Single Instruction Multiple Data (SIMD instruction) method. The mean, variance and waveform factor of each channel data block within the sliding time window are calculated to generate a basic statistical feature set. Based on the basic statistical feature set, the amplitude characteristics of each signal channel within the current sliding time window are calculated, and the corresponding feature timestamps are marked. The amplitude features are correlated with the feature timestamps in a time series to generate structured waveform statistical feature data.

[0040] Specifically, raw vibration signals are acquired by one or more high-frequency vibration sensors deployed at the device's measuring points. After receiving these signals, the high-frequency vibration data is partitioned according to signal channels, such as bearing channels or gearbox channels. Each channel corresponds to an independent storage area, and the partitioned signal data is continuously written to a real-time buffer to form raw waveform data blocks isolated by channel. The real-time buffer adopts a ring buffer structure, supporting high-throughput writing and timestamp marking.

[0041] Parallel waveform statistical calculations are performed on the SIMD instruction set for each channel data block within the real-time buffer. A sliding time window is set for each signal channel, and the window length can be adaptively adjusted according to the signal frequency, typically ranging from 0.1 seconds to 1 second, sliding in fixed steps. At each window position, the mean of multiple sampling points within the window is calculated simultaneously using SIMD instructions. The mean reflects the DC component of the signal, the variance reflects the degree of signal fluctuation, and the waveform factor (the ratio of the effective value to the absolute mean), which is used to distinguish signal types and generate a basic statistical feature set. This parallel computing method can significantly improve the statistical throughput per unit time.

[0042] Based on the basic statistical feature set, the amplitude characteristics of each signal channel within the current sliding time window are calculated, including peak value (maximum absolute value within the window), effective value (root mean square), and kurtosis (normalized value of the fourth-order central moment). A corresponding feature timestamp is marked at the end time of each sliding window; this timestamp is associated with the start and end times of the original data segment within the window, ensuring the temporal uniqueness of the statistical features.

[0043] The calculated amplitude features (peak value, effective value, kurtosis) are time-series correlated with their corresponding feature timestamps, and organized into structured waveform statistical feature data according to channel identifiers and time order. This structured data can be in key-value pair or tabular form, containing fields such as channel ID, timestamp, peak value, effective value, and kurtosis, and is cached in memory for use by subsequent dynamic subscription, trigger judgment, and delay scheduling modules. The original waveform data blocks are retained in a real-time buffer to support subsequent incremental read operations based on timestamp boundaries.

[0044] Step S102: Based on the waveform statistical characteristics, perform dynamic subscription and dynamic unsubscription operations on the signal channel, apply a delay operation to the dynamic unsubscription operation, determine the interception time window, read the target waveform data within the interception time window, and generate an interception segment set. In this optional embodiment, the step of dynamically subscribing to and unsubscribing from the signal channel based on waveform statistical characteristics, applying a delay to the dynamic unsubscribing operation, determining a truncation time window, reading the target waveform data within the truncation time window, and generating a truncation segment set includes: The waveform statistical feature data is compared with the preset trigger conditions in real time, and the signal channel is dynamically subscribed and unsubscribed based on the comparison results. The dynamic unsubscribe operation is delayed and verified through a delayed scheduling mechanism. Based on the timestamp of the dynamic subscription and the timestamp of the dynamic unsubscribe after delay verification, the start and end boundaries of the intercept time window are determined, and the intercept time window is generated. Read the target waveform data within the capture time window, and reconstruct the target waveform data according to the signal channel dimension to generate a capture segment set.

[0045] In this optional embodiment, the step of comparing waveform statistical feature data with preset trigger conditions in real time, and performing dynamic subscription and unsubscription operations on the signal channel based on the comparison results, includes: Set the signal channel to a silent standby state and monitor in real time whether the waveform statistical characteristic data meets the preset trigger conditions; If the preset triggering conditions are met, a dynamic subscription operation is performed on the signal channel, a dynamic subscription event is generated, and the timestamp of the dynamic subscription event is recorded to determine the starting boundary of the interception time window. Based on dynamic subscription events, the operating status of subscribed signal channels is continuously monitored, and waveform statistical feature data is compared with preset unsubscription thresholds in real time to determine whether the operating conditions of the equipment have fallen back to the preset range. If the operating conditions do not fall back to the preset range, the signal channel will be dynamically subscribed again, and the operating status of the subscribed signal channel will be continuously monitored. If the operating conditions fall back to the preset range, the signal channel will be dynamically unsubscribed.

[0046] In this optional embodiment, the step of performing delayed verification of the dynamic unsubscribe operation through a delayed scheduling mechanism, and determining the start and end boundaries of the truncation time window based on the timestamp of the dynamic subscription and the timestamp of the delayed-verified dynamic unsubscribe, and generating the truncation time window includes: Set a delay waiting window for the signal channel that triggers the dynamic unsubscribe operation; The waveform statistical characteristics within the delay waiting window are verified, and the unsubscribe verification result is generated. If the unsubscribe verification result is a dynamic unsubscribe event, then extract the timestamp corresponding to the dynamic unsubscribe event and extract the termination boundary of the time window. If the unsubscribe verification result is not a dynamic unsubscribe event, then cancel the delayed unsubscribe operation, maintain the subscription status of the signal channel, and continue monitoring; The time period is defined by the candidate values ​​of the start and end boundaries, and a time window is generated.

[0047] In this optional embodiment, the step of verifying the waveform statistical characteristics within the delay waiting window and generating an unsubscribe verification result includes: The original waveform data segments are extracted from the delay waiting window and segmented according to the preset time granularity based on the sliding window mechanism to generate a waveform segment dataset. Multi-dimensional statistical feature extraction processing is performed on the waveform segment dataset. The multi-dimensional statistical feature extraction processing includes using the fast Fourier transform algorithm to convert the time domain signal in the waveform segment dataset into the frequency domain signal, extracting the main frequency energy ratio, spectral entropy and frequency band energy distribution features, and generating a multi-dimensional statistical feature vector set. The multi-dimensional statistical feature vector set is compared with the dynamic threshold benchmark to generate a deviation assessment result; The stability score of the deviation evaluation results is calculated using a weighted fusion algorithm, and the unsubscription verification result is generated.

[0048] Specifically, such as Figure 4 As shown, structured waveform statistical feature data, including peak value, RMS value, kurtosis, and corresponding timestamps for each channel, are compared in real time with preset trigger conditions. The preset trigger conditions are set by combining the vibration statistical features under historical normal operating conditions of the equipment with alarm thresholds provided by international standards or equipment manufacturers. For example, for rotating machinery, the preset trigger conditions can be defined as: the peak value exceeds 3 times the average peak value during normal operation, or the RMS value exceeds the upper limit of zone C specified in ISO10816-3, or the kurtosis is greater than 4.

[0049] All signal channels initially operate in a silent standby state, with the system continuously monitoring waveform statistical characteristics to ensure they meet the aforementioned trigger conditions. Once the waveform statistical characteristics of a channel meet the preset trigger conditions, the system immediately performs a dynamic subscription operation on that channel, generating a dynamic subscription event and recording the event's timestamp as a candidate value for the starting boundary of the interception time window. The system continuously monitors the operating status of subscribed signal channels, comparing the waveform statistical characteristic data with a preset unsubscribe threshold range in real time. The preset unsubscribe threshold range is set based on the vibration hysteresis range under normal equipment shutdown or steady-state operating conditions.

[0050] If the operating conditions do not fall back to the preset range, i.e. the vibration is still at a high level, the system will re-subscribe to the channel dynamically and continue monitoring; if the operating conditions fall back to the preset range, the system will trigger the dynamic unsubscribe operation of the signal channel.

[0051] like Figure 5 As shown, a delayed scheduling mechanism is used to perform delayed verification of dynamic unsubscribe operations. Specifically, for the signal channel that triggers the dynamic unsubscribe operation, the system sets a delayed waiting window. The length of the delayed waiting window is determined based on the mechanical time constant of the device and the signal sampling rate. For example, for rolling bearings, the duration of vibration impact decay is approximately 50-200 milliseconds, so the delayed waiting window can be set to 200 milliseconds; for gearboxes, due to their longer transient response, the window can be set to 500 milliseconds. This window length can be configured online by the user according to the device type or automatically optimized through a self-learning algorithm (statistically analyzing the duration distribution of historical valid unsubscribe events).

[0052] Within the delay window, the system verifies the waveform statistical characteristics: it extracts the original waveform data segments from the delay window and segments them according to a preset time granularity based on a sliding window mechanism; the preset time granularity depends on the signal's dominant frequency component and computational load. The system generates a waveform segment dataset; it extracts multi-dimensional statistical features for each segment, including converting the time-domain signal to a frequency-domain signal using a Fast Fourier Transform (FFT), extracting the dominant frequency energy proportion, spectral entropy, and frequency band energy distribution characteristics, and generating a multi-dimensional statistical feature vector set.

[0053] The vector set is compared with a pre-stored dynamic threshold benchmark. The dynamic threshold benchmark is obtained by collecting vibration data during normal, fault-free operation of the equipment and extracting multi-dimensional statistical features, such as the proportion of dominant frequency energy, spectral entropy, and frequency band energy distribution, using the same method. Calculate the deviation between the current value and the benchmark value for each dimension, and generate the deviation assessment results.

[0054] A weighted fusion algorithm is used to calculate the stability score of the deviation assessment results. The weight of each dimension in the weighted fusion algorithm is determined based on the sensitivity of that dimension to the exit condition. The preset reliability threshold is determined by the score distribution of valid cancellation events (i.e., the degree of deviation of statistical characteristics corresponding to the complete shutdown of the equipment) in historical offline test data.

[0055] If the score is higher than the confidence threshold, the unsubscribe verification is considered successful, the dynamic unsubscribe event is confirmed as valid, and the timestamp corresponding to the dynamic unsubscribe event is extracted as the candidate value for the termination boundary of the intercept time window. If the score is lower than the threshold, the delayed unsubscribe operation is canceled, the subscription status of the signal channel is maintained, and monitoring continues. The system uses the timestamp of the dynamic subscription event as the starting boundary candidate value and the timestamp of the dynamic unsubscribe event after delay verification as the termination boundary candidate value to define a time period and generate an intercept time window.

[0056] Based on the start and end boundaries of the generated capture time window, the original waveform data within that time window, i.e., the target waveform data, is quickly located and read from the real-time buffer using a binary positioning algorithm. Then, the read target waveform data is aligned and reconstructed according to the signal channel dimension. For example, consecutive data segments of the same channel are spliced ​​together in chronological order to generate a capture segment set. This capture segment set contains stable vibration data from the start of triggering to the end of unsubscription verification, and has filtered out false trigger segments caused by short-term jitter, for use in subsequent batch learning and coordination steps.

[0057] If the unsubscribe verification fails, the system cancels the current delayed unsubscribe schedule, maintains the original subscription status, and the recorded subscription start boundary remains unchanged. When the working conditions trigger the dynamic unsubscribe operation again, the system restarts the delayed unsubscribe schedule, uses the new unsubscribe trigger timestamp as the candidate value of the termination boundary, and after verification, it determines the interception time window together with the original start boundary.

[0058] The binary search algorithm is a binary search based on ordered timestamp indexes. It utilizes the monotonically increasing timestamp property of data blocks in the buffer to locate the start and end positions of the target time window by continuously halving the search range. The specific operation is shown in Table 1.

[0059] Table 1. Core Flowchart for Signal Channel Dynamic Subscription-Unsubscription and Waveform Capture

[0060] Assuming a measuring point on the high-speed shaft bearing of a wind turbine gearbox, the normal operating conditions show a peak value of 0.5g, an effective value of 0.2g, and a kurtosis of 2.8. The preset trigger condition is a peak value ≥ 1.5g or a kurtosis ≥ 4.0. Initially, the channel remains in silent standby. At a certain moment, the peak value jumps to 2.3g and the kurtosis rises to 5.2, meeting the conditions. The system performs a dynamic subscription, recording the subscription timestamp T0 = 1000ms as the window's starting boundary. The equipment decelerates, the peak value drops back to 0.8g, and the kurtosis decreases to 3.2, falling below the unsubscribe threshold, i.e., 70% of the trigger threshold (peak value < 1.05g and kurtosis < 3.8), triggering a dynamic unsubscribe operation. To prevent jitter, a delay waiting window of 200ms is set, based on the bearing impact attenuation history. Within the window from T0+1500ms to T0+1700ms, waveforms are extracted and segmented in 10ms increments. The frequency domain characteristics are obtained using FFT: the dominant frequency energy percentage decreases from the normal 85% to 62%, the spectral entropy increases from 0.6 to 1.2, and the high-frequency energy (1kHz~5kHz) increases from 6% to 18%. Dynamic threshold benchmark (normal mean ± 3σ): dominant frequency percentage 80% ± 5%, spectral entropy 0.65 ± 0.1, high-frequency energy 6% ± 2%. Deviations are calculated: kurtosis deviation is (5.2-2.8) / 2.8 = 0.86, dominant frequency deviation is (85%-62%) / 85% = 0.27, RMS deviation is (0.8-0.2) / 0.2 = 3.0, and high-frequency energy deviation is (18%-6%) / 6% = 2.0.

[0061] The weighted average (weights: kurtosis 0.4, dominant frequency 0.3, effective value 0.2, high frequency 0.1) yields a stability score of 0.4 × 0.86 + 0.3 × 0.27 + 0.2 × 3.0 + 0.1 × 2.0 = 0.344 + 0.081 + 0.6 + 0.2 = 1.225, which far exceeds the confidence threshold of 0.75. Therefore, the unsubscription is deemed valid, and the unsubscription timestamp T1 = 1700ms is extracted as the window termination boundary. A truncated window [1000ms, 1700ms] is selected, and the waveform is read and reconstructed using binary search to generate a set of truncated segments. Jitter is filtered out to obtain stable vibration segments.

[0062] Step S103: Based on the set of requests to be processed, perform batch learning coordination processing on the truncated fragment set according to the optimal path to obtain batch learning samples; In this optional embodiment, the step of performing batch learning coordination processing on the truncated fragment set according to the optimal path based on the set of requests to be processed, to obtain batch learning samples, includes: Receive several sample interval requests corresponding to each learning task, wherein the sample interval request includes a target time range and a signal channel identifier, and generate a set of requests to be processed. The system summarizes, deduplicates, and merges all sample intervals in the request set to be processed, merging overlapping or adjacent sample intervals into continuous sample intervals to generate a list of read requests. Based on the time order and channel distribution in the read demand list, combined with the physical layout of the storage medium, the optimal read path is calculated and the optimal access plan is generated. According to the optimal access plan, the required sample data is read from the truncated fragments, and the reading results are uniformly saved in memory as a shared sample set to generate batch learning samples.

[0063] In this optional embodiment, the step of calculating the optimal read path and generating the optimal access plan based on the time order and channel distribution in the read demand list, combined with the physical layout of the storage medium, includes: For each read task in the read request list, schedule and sort them according to task priority, time urgency or data size to generate a priority-scheduled read sequence; Based on the priority-scheduled read sequence, read operations of adjacent sample intervals or the same storage channel are merged to generate a merged batch input / output request list; By combining the physical sector layout of the storage medium and the cache hit status, the optimal reading order of the batch input / output request list is calculated, and the optimal access plan is generated.

[0064] Specifically, such as Figure 6 As shown, it receives sample interval requests submitted by multiple learning tasks, such as fault diagnosis model training and health status trend analysis. Each request includes a target time range, such as a start timestamp and an end timestamp, as well as a signal channel identifier, such as a bearing channel or a gearbox channel, and generates a set of requests to be processed.

[0065] All sample intervals in the request set to be processed are summarized, deduplicated, and merged. For example, intervals that overlap or are adjacent in time (with an interval less than a preset threshold) are merged into a continuous sample interval. At the same time, duplicate channels and time periods are removed to generate an optimized list of read requests, thereby reducing the number of subsequent disk I / O operations. The system calculates the optimal read path based on the time order and channel distribution in the read request list, combined with the physical layout of the storage medium, such as the page block structure of a solid-state drive or the cylinder sector of a hard disk drive. First, it schedules and sorts each read task in the read request list according to task priority (e.g., real-time diagnostic tasks take precedence over offline analysis tasks), time urgency (e.g., requests closer to the current time are prioritized), or data size (e.g., small data volumes are prioritized, or large data volumes are processed in batches), generating a priority-based read sequence. Then, based on this read sequence, read operations in adjacent time intervals or belonging to the same storage channel are merged to generate a merged batch I / O request list. Finally, considering the physical sector layout of the storage medium (e.g., avoiding random cross-sector reads and cache hits by prioritizing reads already in the memory cache), the system calculates the optimal read order for this batch I / O request list, minimizing head movement or page access counts, and generating the optimal access plan. According to this optimal access plan, the required sample data is read from the truncated fragment set at once, and the reading results are stored in memory as a shared sample set to generate batch learning samples that can be directly called by multiple learning tasks. Each task calls the shared sample in parallel for training or analysis, and the resources are released in a unified manner after completion, thereby eliminating repeated I / O and repeated computation, and greatly improving the efficiency of multi-task learning.

[0066] Step S104: Write the waveform statistical features, the truncated fragment set, and the batch learning samples into the asynchronous archiving queue and generate an archive file; wherein, the data block index in the archive file determines the original data block range; Specifically, the waveform statistical features generated during the online processing phase, which include the peak value, effective value, kurtosis and corresponding timestamp of each channel, the set of truncated segments generated during the runtime control phase, and the batch learning samples generated during the batch learning coordination phase are uniformly encapsulated into archive events. These events are written to an asynchronous queue in chronological order. The real-time main thread continues to execute subsequent processing immediately after writing to the queue, without waiting for the data to be written to disk, thereby avoiding blocking the real-time link.

[0067] An independent background thread asynchronously consumes archived events in the queue: For each event, the system directly converts the waveform data array, i.e., the original sampling point sequence, into a byte sequence via memory mapping, skipping the overhead of traditional serialization format conversion and achieving zero-copy writing. During the writing process, the byte sequence is appended to the archive file according to a fixed data block size, and a data block index is generated for each data block. This index records three parts of information: the start timestamp, the end timestamp, and the physical offset position in the archive file. This index information can be stored in an inverted index table in memory or in a separate index file, thereby determining the interval of each original data block, i.e., the time range from the start timestamp to the end timestamp. The archive file ultimately consists of multiple such indexed data blocks. During subsequent offline review, the index can be used to quickly locate the data block corresponding to the target time range and its physical location, achieving efficient temporal semantic reading.

[0068] Step S105: In the offline stage, the intersection and truncation operation is performed on the target truncation interval and the original data block interval determined by the data block index in the archive file to generate an offline fragment file for offline review, export and re-analysis.

[0069] In this optional embodiment, the step of performing an intersection truncation operation on the target truncation interval and the original data block interval determined by the data block index in the archive file during the offline stage to generate an offline fragment file for offline review, export, and re-analysis includes: Receive the target cut-off range input by the user, and retrieve the original data block range and data block index that overlap with the target cut-off range in the archive file to generate the block original dataset to be processed; Determine whether each original data block in the original dataset to be processed overlaps with the target time range, filter out non-overlapping data blocks, extract valid data segments with overlap, and generate a list of filtered valid data segments. The data segments in the filtered list of valid data segments are sorted and concatenated in chronological order, and timestamps are added at the junctions of adjacent data segments to generate calibrated continuous time-series data. The calibrated continuous time series data is packaged into a standard offline file format to generate offline fragment files for offline verification, export, and reanalysis.

[0070] Specifically, during the offline review phase, the system receives the target intercept interval input by the user. This target intercept interval includes a start time and an end time. For example, the user selects a time range before and after a certain fault occurs through the software interface.

[0071] Based on the time boundary of the target interval, the system retrieves all possible overlapping original data blocks from the archive file: by traversing each original data block interval recorded in the data block index, such as the start and end timestamps and the corresponding physical offset positions, the system filters out the data blocks that intersect with the target interval in time, and generates the block-based original dataset to be processed.

[0072] For each original data block in the dataset, perform an intersection operation between the time interval it covers and the target truncation interval: determine whether there is an overlap between the two intervals. If there is no overlap, filter and discard the data block; if there is an overlap, calculate the start and end times of the overlapping sub-interval, and accurately locate and read the waveform data corresponding to the overlapping sub-interval from the archive file according to the physical offset position in the data block index, and extract it as a valid data segment.

[0073] All extracted valid data segments are aggregated to generate a filtered list of valid data segments. The system sorts these valid data segments according to their chronological order and concatenates adjacent segments. Since there may be slight time intervals or overlaps at the concatenation points due to data block boundaries, the system adds timestamps during concatenation, such as recording the original timestamp range of each data segment, forming calibrated continuous time-series data to ensure no loss or misalignment of temporal semantics. The system then encapsulates this calibrated continuous time-series data into standard offline file formats, such as CSV, MAT, and Parquet, generating offline fragment files that are fully aligned with the online time-series data. These fragments are available for users to perform offline review, export data, or reanalyze fault causes.

[0074] Figure 2 An embodiment of a high-frequency vibration signal collaborative processing system of the present invention is shown.

[0075] In this optional embodiment, the high-frequency vibration signal collaborative processing system includes: Signal receiving module 201 is used to receive high-frequency vibration signals and write them into the real-time buffer according to the signal channel, perform waveform statistical calculations on the real-time buffer, and generate waveform statistical features. The waveform calculation module 202 is used to perform dynamic subscription and dynamic unsubscription operations on the signal channel based on the waveform statistical characteristics, apply a delay operation to the dynamic unsubscription operation, determine the truncation time window, read the target waveform data within the truncation time window, and generate a set of truncation segments. The learning coordination module 203 is used to perform batch learning coordination processing on the truncated fragment set according to the optimal path based on the set of requests to be processed, so as to obtain batch learning samples. The feature summarization module 204 is used to write waveform statistical features, truncated fragment sets and batch learning samples into an asynchronous archiving queue and generate an archive file; wherein, the data block index in the archive file determines the original data block range; The offline management module 205 is used to perform an intersection truncation operation on the target truncation interval and the original data block interval determined by the data block index in the archive file during the offline stage, and generate an offline fragment file for offline review, export and re-analysis.

[0076] To facilitate understanding of the above technical solutions of the present invention, the following further explains the above technical solutions of the present invention from the perspective of architecture and principle, as follows: High-frequency vibration signals are written to real-time buffers by channel. SIMD parallel computation is used to calculate waveform statistical features, such as peak value, RMS value, and kurtosis. Incremental reading is achieved using binary positioning based on timestamp boundaries, avoiding resource contention caused by full data transfer. Based on the comparison results of waveform statistical features and preset thresholds, dynamic subscription and unsubscription are performed on the signal channels. A cancelable delay scheduling mechanism is used to perform delay verification of the unsubscription operation. A stable, jitter-resistant truncation time window is generated, with the subscription timestamp as the starting boundary and the delayed-verified unsubscription timestamp as the ending boundary. Target waveform data is read from this window and reconstructed into a set of truncation segments.

[0077] It receives sample interval requests from multiple learning tasks, optimizes the reading path through request merging, priority scheduling, and storage layout, and reads sample data in batches using a shared loading method to generate a shared sample set, eliminating redundant I / O and computational overhead. Waveform statistical features, truncated fragment sets, and batch learning samples are encapsulated as archive events and written to an asynchronous queue. A background thread then writes these to disk using byte mapping and establishes a data block index containing timestamp intervals and physical offsets, thereby reducing the risk of real-time link congestion.

[0078] During the offline phase, the target truncation range input by the user is intersected with the original data block range in the archived data block index one by one to extract overlapping valid data segments. After time-series splicing and timestamp calibration, an offline fragment file that is completely aligned with the online version is generated, supporting accurate review and re-analysis.

[0079] This invention integrates data acquisition, trigger control, intelligent learning, efficient archiving, and offline retrieval, solving problems such as resource contention, false triggering, high overhead, and inconsistent time semantics.

[0080] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores static and dynamic information data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0081] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0082] In addition, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0083] In addition, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0084] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0085] This invention is not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this invention is limited only by the appended claims.

Claims

1. A method for collaborative processing of high-frequency vibration signals, characterized in that, include: The system receives high-frequency vibration signals and writes them into a real-time buffer according to the signal channel. It then performs waveform statistical calculations on the real-time buffer to generate waveform statistical features. Based on the waveform statistical characteristics, the signal channel is dynamically subscribed and unsubscribed, and a delay operation is applied to the dynamic unsubscription operation to determine the interception time window, read the target waveform data within the interception time window, and generate an interception segment set. Based on the set of requests to be processed, the truncated fragment set is processed in batch learning coordination according to the optimal path to obtain batch learning samples. The waveform statistical features, the truncated fragment set, and the batch learning samples are written into an asynchronous archiving queue and an archive file is generated; wherein, the data block index in the archive file determines the original data block range; During the offline phase, an intersection truncation operation is performed on the target truncation interval and the original data block interval determined by the data block index in the archive file to generate an offline fragment file for offline review, export, and reanalysis.

2. The high-frequency vibration signal collaborative processing method according to claim 1, characterized in that, The process of receiving high-frequency vibration signals and writing them into a real-time buffer according to signal channels, and performing waveform statistical calculations on the real-time buffer to generate waveform statistical features includes: High-frequency vibration signals are collected, the high-frequency vibration signals are divided into partitions according to signal channels, and the partitioned signal data is stored in a real-time buffer to form channel data blocks. By using Single Instruction Multiple Data to perform waveform statistical calculations on each channel data block in the real-time buffer in parallel, the mean, variance and waveform factor of each channel data block are calculated within the sliding time window to generate a basic statistical feature set. Based on the basic statistical feature set, the amplitude characteristics of each signal channel within the current sliding time window are calculated, and the corresponding feature timestamps are marked. The amplitude features are correlated with the feature timestamps in a time series to generate structured waveform statistical feature data.

3. The high-frequency vibration signal collaborative processing method according to claim 1, characterized in that, The process of dynamically subscribing to and unsubscribing from the signal channel based on waveform statistical characteristics, applying a delay to the dynamic unsubscribing operation, determining the truncation time window, reading the target waveform data within the truncation time window, and generating a set of truncation segments includes: The waveform statistical feature data is compared with the preset trigger conditions in real time, and the signal channel is dynamically subscribed and unsubscribed based on the comparison results. The dynamic unsubscribe operation is delayed and verified through a delayed scheduling mechanism. Based on the timestamp of the dynamic subscription and the timestamp of the dynamic unsubscribe after delay verification, the start and end boundaries of the intercept time window are determined, and the intercept time window is generated. Read the target waveform data within the capture time window, and reconstruct the target waveform data according to the signal channel dimension to generate a capture segment set.

4. The high-frequency vibration signal collaborative processing method according to claim 3, characterized in that, The step of comparing waveform statistical feature data with preset trigger conditions in real time, and performing dynamic subscription and unsubscription operations on the signal channel based on the comparison results, includes: Set the signal channel to a silent standby state and monitor in real time whether the waveform statistical characteristic data meets the preset trigger conditions; If the preset triggering conditions are met, a dynamic subscription operation is performed on the signal channel, a dynamic subscription event is generated, and the timestamp of the dynamic subscription event is recorded to determine the starting boundary of the interception time window. Based on dynamic subscription events, the operating status of subscribed signal channels is continuously monitored, and waveform statistical feature data is compared with preset unsubscription thresholds in real time to determine whether the operating conditions of the equipment have fallen back to the preset range. If the operating conditions do not fall back to the preset range, the signal channel will be dynamically subscribed again, and the operating status of the subscribed signal channel will be continuously monitored. If the operating conditions fall back to the preset range, the signal channel will be dynamically unsubscribed.

5. The high-frequency vibration signal collaborative processing method according to claim 4, characterized in that, The process of delaying and verifying dynamic unsubscription operations through a delayed scheduling mechanism, and determining the start and end boundaries of the truncation time window based on the timestamp of the dynamic subscription and the timestamp of the delayed-verified dynamic unsubscription, and generating the truncation time window includes: Set a delay waiting window for the signal channel that triggers the dynamic unsubscribe operation; The waveform statistical characteristics within the delay waiting window are verified, and the unsubscribe verification result is generated. If the unsubscribe verification result is a dynamic unsubscribe event, then extract the timestamp corresponding to the dynamic unsubscribe event and extract the termination boundary of the time window. If the unsubscribe verification result is not a dynamic unsubscribe event, then cancel the delayed unsubscribe operation, maintain the subscription status of the signal channel, and continue monitoring; The time period is defined by the candidate values ​​of the start and end boundaries, and a time window is generated.

6. The high-frequency vibration signal collaborative processing method according to claim 5, characterized in that, The step of verifying the waveform statistical characteristics within the delay waiting window and generating the unsubscribe verification result includes: The original waveform data segments are extracted from the delay waiting window and segmented according to the preset time granularity based on the sliding window mechanism to generate a waveform segment dataset. Multi-dimensional statistical feature extraction processing is performed on the waveform segment dataset. The multi-dimensional statistical feature extraction processing includes using the fast Fourier transform algorithm to convert the time domain signal in the waveform segment dataset into the frequency domain signal, extracting the main frequency energy ratio, spectral entropy and frequency band energy distribution features, and generating a multi-dimensional statistical feature vector set. The multi-dimensional statistical feature vector set is compared with the dynamic threshold benchmark to generate a deviation assessment result; The stability score of the deviation evaluation results is calculated using a weighted fusion algorithm, and the unsubscription verification result is generated.

7. The high-frequency vibration signal collaborative processing method according to claim 1, characterized in that, Based on the set of requests to be processed, the truncated fragment set is batch-learned and coordinated according to the optimal path to obtain batch learning samples, including: Receive several sample interval requests corresponding to each learning task, wherein the sample interval request includes a target time range and a signal channel identifier, and generate a set of requests to be processed. The system summarizes, deduplicates, and merges all sample intervals in the request set to be processed, merging overlapping or adjacent sample intervals into continuous sample intervals to generate a list of read requests. Based on the time order and channel distribution in the read demand list, combined with the physical layout of the storage medium, the optimal read path is calculated and the optimal access plan is generated. According to the optimal access plan, the required sample data is read from the truncated fragments, and the reading results are uniformly saved in memory as a shared sample set to generate batch learning samples.

8. The high-frequency vibration signal collaborative processing method according to claim 7, characterized in that, The step of calculating the optimal read path and generating the optimal access plan based on the time order and channel distribution in the read request list, combined with the physical layout of the storage medium, includes: For each read task in the read request list, schedule and sort them according to task priority, time urgency or data size to generate a priority-scheduled read sequence; Based on the priority-scheduled read sequence, read operations of adjacent sample intervals or the same storage channel are merged to generate a merged batch input / output request list; By combining the physical sector layout of the storage medium and the cache hit status, the optimal reading order of the batch input / output request list is calculated, and the optimal access plan is generated.

9. The high-frequency vibration signal collaborative processing method according to claim 1, characterized in that, In the offline phase, the intersection and truncation operation is performed on the target truncation interval and the original data block interval determined by the data block index in the archive file to generate an offline fragment file for offline review, export, and reanalysis, including: Receive the target cut-off range input by the user, and retrieve the original data block range and data block index that overlap with the target cut-off range in the archive file to generate the block original dataset to be processed; Determine whether each original data block in the original dataset to be processed overlaps with the target time range, filter out non-overlapping data blocks, extract valid data segments with overlap, and generate a list of filtered valid data segments. The data segments in the filtered list of valid data segments are sorted and concatenated in chronological order, and timestamps are added at the junctions of adjacent data segments to generate calibrated continuous time-series data. The calibrated continuous time series data is packaged into a standard offline file format to generate offline fragment files for offline verification, export, and reanalysis.

10. A high-frequency vibration signal collaborative processing system, characterized in that, include: The signal receiving module is used to receive high-frequency vibration signals and write them into the real-time buffer according to the signal channel, and to perform waveform statistical calculations on the real-time buffer to generate waveform statistical features. The waveform calculation module is used to dynamically subscribe to and unsubscribe from the signal channel based on the waveform statistical characteristics, apply a delay to the dynamic unsubscribe operation, determine the truncation time window, read the target waveform data within the truncation time window, and generate a set of truncation segments. The learning coordination module is used to perform batch learning coordination processing on the truncated fragment set according to the optimal path based on the set of requests to be processed, so as to obtain batch learning samples. The feature summarization module is used to write waveform statistical features, truncated fragment sets, and batch learning samples into an asynchronous archiving queue and generate an archive file; wherein the data block index in the archive file determines the original data block range; The offline management module is used to perform intersection and truncation operations on the target truncation range and the original data block range determined by the data block index in the archive file during the offline stage, generating offline fragment files for offline review, export, and re-analysis.

Images