Wind turbine generator state monitoring method based on edge calculation
By directly analyzing the kurtosis values of wind turbine vibration signals through edge computing nodes and dynamically adjusting the monitoring task frequency and processor time slice allocation, the problem of network bandwidth limitation in traditional wind turbine condition monitoring is solved, enabling efficient fault early warning and diagnosis and improving the utilization efficiency of computing resources.
Patent Information
- Application Number
- CN202511611859.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-05
- Publication Date
- 2026-02-03
AI Technical Summary
Traditional wind turbine condition monitoring methods rely on long-distance data transmission to a central data center, resulting in limited or unstable network bandwidth. This affects the timeliness of fault warnings and the effective allocation of computing resources. It is difficult to dynamically adjust the allocation of computing resources according to real-time changes in equipment status, leading to wasted computing resources and insufficient depth of fault diagnosis.
A wind turbine condition monitoring method based on edge computing is adopted. The kurtosis value of the vibration signal is directly analyzed through edge computing nodes, the activation frequency of the monitoring task is dynamically adjusted, a fast Fourier transform is performed to identify the center frequency of the main frequency band, and the allocation of processor time slices is adjusted according to the frequency.
This has enabled a shift from periodic monitoring to state-driven monitoring, improving the response speed and analysis depth of fault warnings, enhancing the accuracy of computing resource allocation and utilization efficiency, and ensuring a responsive approach to operational and maintenance needs.
Smart Images

Figure CN121456657A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of resource allocation, in particular to a wind turbine state monitoring method based on edge computing. BACKGROUND
[0002] The technical field of resource allocation is a core component of computer system management, and its core matters include computing power scheduling, storage space management, and network bandwidth allocation. This technical field uses the scheduling strategy of the operating system kernel, the virtualization management program, or the distributed computing framework to dynamically or statically assign processor time, memory blocks, and communication links to multiple concurrent computing tasks, and is widely used in server clusters, cloud computing platforms, and data center operations.
[0003] Among them, the traditional wind turbine state monitoring method refers to the data collection and analysis process for evaluating the operational health of wind turbines. Its technical matter is to achieve early fault warning for key components such as gearboxes and bearings. This method usually relies on vibration sensors and temperature sensors deployed in the cabin to collect time series data, and transmits all raw data or preliminary aggregated data to a central data center through wired or wireless networks. In the central server, Fourier transform is performed for spectral analysis or statistical process control charts are constructed to identify abnormal patterns.
[0004] In practical application, the traditional wind turbine state monitoring method relies on transmitting all raw or preliminary aggregated data to a central data center for centralized analysis. In an environment with limited or unstable network bandwidth, data congestion and transmission delays are likely to occur, affecting the timeliness of fault warning. Especially when a sudden fault occurs in the wind turbine, this inherent data transmission bottleneck may prevent the capture of critical early abnormal signals. In addition, the scheduling strategy of the computing task executed on the central server is relatively fixed, lacking awareness of different fault types and their urgency, making it difficult to dynamically adjust the allocation of computing resources according to real-time changes in the state of the front-end device. For example, the system may allocate the same processor time for spectral analysis regardless of whether the device is running smoothly or experiencing minor abnormalities. This not only wastes computing resources, but also fails to concentrate computing power for deeper diagnosis when fault characteristics are significant, thereby reducing overall operational efficiency. SUMMARY
[0005] The present application aims to solve the shortcomings in the prior art and proposes a wind turbine state monitoring method based on edge computing.
[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solution: a wind turbine state monitoring method based on edge computing, comprising the following steps: S1: Obtain a data block formed by a wind turbine vibration acceleration signal collected by an edge computing node, and calculate a time domain kurtosis value of the data block; S2: Obtain a preset kurtosis warning line, a kurtosis alarm line and a standard scheduling period in an operating system kernel, compare the time domain kurtosis value with the kurtosis warning line and the kurtosis alarm line, determine a scheduling mode of a monitoring task as the standard scheduling period, half of the standard scheduling period or immediate activation, and generate a monitoring task activation instruction; S3: Based on the monitoring task activation instruction, call a fast Fourier transform to process the data block, generate a frequency component and a corresponding energy intensity value, construct a frequency dimension representation, and identify a frequency of the frequency component with the highest energy intensity value as a main frequency band center frequency from the frequency dimension representation; S4: Obtain a frequency-time slice mapping table in the operating system kernel, query a processor time slice length retrieved from the frequency-time slice mapping table based on the main frequency band center frequency, and adjust the processor time slice length as an upper limit of an execution time of the monitoring task.
[0007] As a further scheme of the present application, the data block includes a sample point sequence, a sampling frequency and a collection timestamp, the monitoring task activation instruction specifically is a task priority, a scheduling timestamp and a target task identifier, the frequency dimension representation includes a frequency resolution, an energy spectrum vector and a frequency index list, and the frequency-time slice mapping table includes a fault characteristic frequency range, a normal working frequency range and a corresponding time slice quota.
[0008] As a further scheme of the present application, the step of S1 specifically is: Obtain a data block formed by a wind turbine vibration acceleration signal collected by an edge computing node, and calculate a time domain mean value and a time domain standard deviation of a sample point sequence included in the data block; Based on the time domain mean value and the time domain standard deviation, calculate a fourth-order central moment of the sample point sequence; Calculate a ratio of the fourth-order central moment and a fourth power of the time domain standard deviation to obtain a time domain kurtosis value of the data block.
[0009] As a further scheme of the present application, the step of S2 specifically is: Obtain a preset kurtosis warning line, a kurtosis alarm line and a standard scheduling period in an operating system kernel; Compare the time domain kurtosis value with the kurtosis warning line and the kurtosis alarm line, and determine whether the time domain kurtosis value is less than the kurtosis warning line, if yes, determine the scheduling mode as the standard scheduling period; determining whether the time-domain kurtosis value is greater than or equal to the kurtosis warning line and less than the kurtosis alarm line, if yes, determining the scheduling mode as half of the standard scheduling period; determining whether the time-domain kurtosis value is greater than or equal to the kurtosis alarm line, if yes, determining the scheduling mode as immediate activation; generating a monitoring task activation instruction based on the determined scheduling mode.
[0010] As a further scheme of the present application, the step S3 specifically includes: based on the monitoring task activation instruction, calling fast Fourier transform to process the sampling point sequence in the data block to obtain a frequency component and a corresponding energy intensity value; combining a sampling frequency and a length of the sampling point sequence to calculate a frequency resolution, and constructing a frequency dimension representation including the frequency resolution, the energy spectrum vector and the frequency index list; traversing the energy spectrum vector to identify the frequency component with the highest energy intensity value; extracting a frequency corresponding to the frequency component in the frequency index list as a main frequency band center frequency.
[0011] As a further scheme of the present application, the step S4 specifically includes: obtaining a frequency-time slice mapping table in an operating system kernel; based on the main frequency band center frequency, querying the frequency-time slice mapping table to match a fault characteristic frequency range or a normal working frequency range to which the main frequency band center frequency belongs; retrieving a time slice quota corresponding to the matched fault characteristic frequency range or normal working frequency range, and determining the time slice quota as a processor time slice length; adjusting the processor time slice length as an upper limit of execution time of a monitoring task.
[0012] As a further scheme of the present application, the time-domain kurtosis value is calculated by the following formula: ; wherein, is a total number of sampling points in the sampling point sequence, is a vibration acceleration value of an i-th sampling point in the sampling point sequence, is a time-domain mean value of the sampling point sequence.
[0013] As a further scheme of the present application, the generation process of the monitoring task activation instruction further includes dynamically determining the task priority according to the scheduling mode. obtaining a preset standard priority, a preset medium priority and a preset highest priority in the operating system kernel; when the scheduling mode is the standard scheduling period, setting the task priority as the standard priority; when the scheduling mode is half of the standard scheduling period, setting the task priority as the medium priority; when the scheduling mode is immediate activation, setting the task priority as the highest priority.
[0014] As a further scheme of the present application, after the frequency component with the highest energy intensity value is identified, further comprising: obtaining an average value of all the energy intensity values in the energy spectrum vector, calculating an energy average value; obtaining a preset energy threshold factor in the operating system kernel; calculating a ratio of the highest energy intensity value to the energy average value, obtaining an energy peak value ratio; judging whether the energy peak value ratio is greater than the energy threshold factor, if yes, confirming that the frequency component is an effective fault frequency component, and extracting the frequency corresponding to the effective fault frequency component as the main frequency band center frequency; if the energy peak value ratio is not greater than the energy threshold factor, setting the main frequency band center frequency as a null value or a preset normal frequency value.
[0015] As a further scheme of the present application, the adjustment process of the processor time slice length is specifically: obtaining the basic time slice quota searched in the frequency-time slice mapping table ; obtaining the time domain kurtosis value and the kurtosis warning line ; calculating the processor time slice length by the following formula : ; wherein, is a preset time slice adjustment sensitivity coefficient in the operating system kernel, used for adjusting the influence degree of the time domain kurtosis value exceeding the kurtosis warning line on the processor time slice length.
[0016] Compared with the prior art, the present application has the following advantages and positive effects: In the present application, by directly analyzing the kurtosis value of the vibration signal at the edge computing node, the activation frequency of the monitoring task can be dynamically adjusted according to the real-time health status of the unit, abandoning the fixed cycle inspection mode, realizing the transformation from periodic monitoring to state-driven monitoring, and when potential fault signs are identified, the data block is subjected to fast Fourier transform, and the strongest vibration main frequency band is extracted, and the center frequency of the main frequency band becomes the direct basis for adjusting the processor time slice allocation in the operating system kernel, assigning the subsequent fine fault diagnosis task with computing resources matching its complexity, not only significantly improving the accuracy and utilization efficiency of computing resource allocation, but also ensuring the response speed and analysis depth of fault warning, achieving efficient scheduling of computing power and sensitive response to operation and maintenance needs. BRIEF DESCRIPTION OF DRAWINGS
[0017] Figure 1 The present application provides a wind turbine state monitoring method based on edge computing, and the flow chart of the method is shown in the figure. Figure 2 The present application provides a time domain kurtosis value calculation flow chart. Figure 3 The present application provides a monitoring task scheduling mode and priority determination flow chart. Figure 4 The present application provides a main frequency band center frequency identification flow chart. Figure 5 The present application provides a processor time slice length adjustment flow chart. DETAILED DESCRIPTION
[0018] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme realized by software will be described in detail below combined with system architecture diagram and embodiment. It should be understood that the specific embodiments described herein are only used to explain the technical scheme of the present application, and do not constitute a limitation on the scope of protection.
[0019] In the description of the present application, the system architecture relationship or data processing flow indicated by the terms "level", "module", "interface", "data flow", "client", "server" and the like are defined based on the corresponding architecture diagram or flow chart of the embodiment. This way of expression is only used to clearly explain the logical relationship of each element in the technical scheme, and is not limited to the physical deployment form. The "multiple" contains two or more technical units, including but not limited to multiple data nodes, processing threads, service instances or functional components, etc. Extensible elements, the specific number is determined according to the actual business scenario.
[0020] Please refer to Figure 1 and Figure 2 The present application provides a technical scheme: a wind turbine state monitoring method based on edge computing, comprising the following steps: S1: Obtain the data block formed by the vibration acceleration signal of the wind turbine collected by the edge computing node, and calculate the time domain kurtosis value of the data block; The data block includes the sampling point sequence, sampling frequency, and acquisition timestamp; The specific steps of S1 are as follows: The data block is formed by acquiring the vibration acceleration signal of the wind turbine collected by the edge computing node, and the time domain mean and time domain standard deviation of the sampling point sequence included in the data block are calculated. The fourth central moment of the sampling point sequence is calculated based on the time domain mean and time domain standard deviation. The time-domain kurtosis value of the data block is obtained by calculating the ratio of the fourth central moment to the fourth power of the time-domain standard deviation. Time domain kurtosis value Calculated using the following formula: ; in, The total number of sampling points in the sampling point sequence. The sampling point sequence is the first Vibration acceleration values at each sampling point This represents the time-domain mean of the sample point sequence.
[0021] The edge computing node, specifically an industrial PC (Advantech ARK-2250L) installed in the main control cabinet within the nacelle of a wind turbine (such as the Vestas V112-3.0MW unit), acquires vibration acceleration signals from a piezoelectric accelerometer (PCB352C33) deployed on the high-speed stage bearing housing of the gearbox (monitoring point number ZDC-001). This signal is then transmitted via a data acquisition card (NI-9234) to... The sampling frequency of Hz is used for analog-to-digital conversion, and data blocks are formed continuously.
[0022] exist At a given time (e.g., UNIX timestamp 1761657600, corresponding to 2025-10-28T06:40:00Z), retrieve a data block containing: Sampling point sequence: a sequence containing A floating-point array of sampling points, which is generated by the acquisition card within 1 second ( The collected vibration acceleration values Composition. For ease of explanation, an example is provided below: .
[0023] Sampling frequency: Hz.
[0024] Data collection timestamp: 1761657600.
[0025] First, calculate the data block. Time-domain mean of each sampling point sequence This calculation process involves combining all sequences... indivual Numerical value (from) arrive Accumulate the results and then divide by the total number of sampling points. In this embodiment, for sampling points (Data sequence omitted) Perform calculations to obtain the time-domain mean. .
[0026] Next, calculate the time-domain standard deviation (or its square, i.e., variance). This process involves sampling each point in the sequence. Calculate its value and the mean. The difference Then calculate the square of the difference. All Sum the squared values, then divide by . .
[0027] In this embodiment, this value is... .
[0028] Subsequently, based on the time domain mean The fourth central moment of the sampling point sequence is calculated using the time-domain standard deviation. This process involves: for each sampling point in the sequence... Calculate its value and the mean. The difference Then calculate the fourth power of the difference. All The fourth power values are summed, and then divided by . .
[0029] In this embodiment, this value is... .
[0030] Finally, the time-domain kurtosis value of the data block is obtained by calculating the ratio of the fourth-order central moment to the fourth power of the time-domain standard deviation (i.e., the square of the variance). .
[0031] The time domain kurtosis value Calculation formula middle: This represents the final calculated kurtosis value. This represents the total number of sampling points in this example. . The index of the sampling point sequence, incrementing from 1 to... . For the first in the sequence Measured vibration acceleration values at each sampling point. The time-domain mean of the sequence, in this example . Indicates to arrive Summing all terms.
[0032] The numerator of the formula The calculation is for the fourth central moment, which is the average of the fourth power of the deviation of each point in the sequence from the mean.
[0033] The denominator of the formula It calculates the square of the variance (second central moment), which is the fourth power of the time-domain standard deviation.
[0034] The entire formula operates by normalizing the degree of “peaks” and “tails” in the signal distribution by calculating the ratio of the fourth-order central moment to the square of the variance.
[0035] The advantage of the formula lies in its ability to quantify the impact or outlier sensitivity of a signal by comparing the fourth-order central moment with the square of the variance. Kurtosis values are particularly sensitive to transient impact faults such as early pitting and cracking in wind turbine gears or bearings; a significant increase in the value (e.g., from around 3 to 6 or higher) indicates the onset of a fault.
[0036] Substitute the calculated values into the formula: Molecular (fourth central moment) = .
[0037] Denominator (squared variance) = .
[0038] .
[0039] The result (Dimensionless) is the time-domain kurtosis value finally obtained in step S1. This value is close to 3.0, indicating that the vibration signal distribution of the current data block is close to a Gaussian distribution, indicating that the wind turbine is in a healthy operating state.
[0040] Please see Figure 1 and Figure 3 S2: Obtain the preset kurtosis warning threshold, kurtosis alarm threshold and standard scheduling cycle in the operating system kernel. Compare the time domain kurtosis value with the kurtosis warning threshold and kurtosis alarm threshold to determine whether the scheduling mode of the monitoring task is the standard scheduling cycle, half of the standard scheduling cycle or immediate activation, and generate a monitoring task activation command. The monitoring task activation command specifically includes the task priority, scheduling timestamp, and target task identifier; The specific steps of S2 are as follows: Obtain the preset kurtosis warning threshold, kurtosis alarm threshold, and standard scheduling cycle in the operating system kernel; Compare the time domain kurtosis value with the kurtosis warning line and the kurtosis alarm line to determine whether the time domain kurtosis value is less than the kurtosis warning line. If so, determine that the scheduling mode is the standard scheduling cycle. Determine whether the time domain kurtosis value is greater than or equal to the kurtosis warning line and less than the kurtosis alarm line. If so, determine that the scheduling mode is half of the standard scheduling cycle. Determine if the time-domain kurtosis value is greater than or equal to the kurtosis alarm threshold; if so, determine the scheduling mode as immediate activation. Based on the determined scheduling method, generate a monitoring task activation instruction; The process of monitoring the generation of task activation instructions also includes dynamically determining task priorities based on the scheduling method; Obtain the preset standard priority, medium priority, and highest priority in the operating system kernel; When the scheduling mode is the standard scheduling cycle, the task priority is set to the standard priority; When the scheduling mode is half of the standard scheduling cycle, set the task priority to medium priority; When the scheduling mode is set to immediate activation, the task priority is set to the highest priority.
[0041] The preset kurtosis warning line is obtained from a memory-specific configuration region of the operating system (a trimmed real-time Linux kernel) of the edge computing node (Advantech ARK-2250L) (e.g., from the / etc / edge_monitor.conf file loaded at boot). , ravine warning line With standard scheduling cycle .
[0042] and The settings are based on statistical analysis of operating data of a wind turbine (model Vestas V112-3.0MW) under different health conditions. To determine these two values, the following experimental verification process was performed: 1. Data Acquisition Phase: Vibration data were continuously collected under three distinct conditions (State A: Newly installed unit, healthy; State B: After 1 year of operation, early wear of the gearbox was observed, confirmed by endoscopic examination; State C: After 3 years of operation, the high-speed bearing exhibited a peeling failure, producing obvious abnormal noise).
[0043] 2. Data Processing: In each state, 2000 data blocks as described in S1 are collected ( ), and calculate the kurtosis value for each data block. .
[0044] 3. Statistical Analysis: The kurtosis values under the three states were statistically analyzed, and the results are shown in Table 1.
[0045] Table 1. Experimental data on kurtosis threshold setting As shown in Table 1, the kurtosis values of the healthy state are mainly distributed below 4.40. To ensure sensitivity to early wear while avoiding normal fluctuations in the healthy state ( This could lead to misjudgment and cause the ravine warning line to be misjudged. Set as .
[0046] The distributions of states B and C overlap, but the mean of state B is 4.80, while the mean of state C is 7.20. To effectively distinguish between "wear" and "failure," a value between their means is selected as the alarm threshold, set to... .
[0047] Standard scheduling cycle Set as Seconds (10 minutes) are the fixed intervals for routine condition checks during normal unit operation.
[0048] After obtaining the above preset values: , , Second.
[0049] The calculation obtained in S1 and and Compare them.
[0050] First, determine Is (3.0976) less than (4.5).
[0051] The comparison result is "yes".
[0052] Therefore, the scheduling method is determined to be the standard scheduling period, i.e. Second.
[0053] (like : judge The answer is "No".
[0054] Next, make a judgment (5.1) Is it greater than or equal to? (4.5) and less than (6.0).
[0055] and The comparison result is "yes".
[0056] The scheduling method is determined to be half of the standard scheduling cycle. Second.
[0057] like : judge The answer is "No".
[0058] judge and The answer is "No".
[0059] Next, make a judgment (6.8) Is it greater than or equal to? (6.0).
[0060] The comparison result is "yes".
[0061] The scheduling method is set to immediate activation. Second.) Based on the scheduling method (standard scheduling cycle, 600 seconds) determined in this embodiment, a monitoring task activation command is generated.
[0062] This process also includes retrieving the preset priority value from the operating system kernel: Standard priority (P_std): 30 (Linux real-time priority, the smaller the value, the higher the priority) Medium priority (P_mid): 20 Highest priority (P_high): 10 When the scheduling method is the standard scheduling cycle, the task priority will be... Set to standard priority . .
[0063] (If the scheduling method is 300 seconds, then) If activated immediately, then 。 The instruction also includes a scheduling timestamp. and target task identifier .
[0064] Fixed as "TASK_FFT_ANLYS".
[0065] By obtaining the current timestamp (e.g., 1761657601) and a defined scheduling period (600 seconds) added together to get the result.
[0066] .
[0067] (If activated immediately, Set to 0 or 。 Finally, the monitoring task activation command generated by S2 is: {Task priority: 30, scheduling timestamp: 1761658201, target task identifier: "TASK_FFT_ANLYS"}.
[0068] Please see Figure 1 and Figure 4 S3: Based on the monitoring task activation command, call the fast Fourier transform to process the data block, generate frequency components and corresponding energy intensity values, construct a frequency dimension representation, and identify the frequency component with the highest energy intensity value from the frequency dimension representation as the center frequency of the main frequency band. The frequency dimension representation includes frequency resolution, energy spectrum vector, and frequency index list; The specific steps for S3 are as follows: Based on the monitoring task activation command, the fast Fourier transform is invoked to process the sampling point sequence in the data block to obtain the frequency components and the corresponding energy intensity values. The frequency resolution is calculated by combining the sampling frequency and the length of the sampling point sequence, and a frequency dimension representation including the frequency resolution, energy spectrum vector, and frequency index list is constructed. Traverse the energy spectrum vector to identify the frequency component with the highest energy intensity value; Extract the frequency component corresponding to the frequency index list and use it as the center frequency of the main frequency band; After identifying the frequency component with the highest energy intensity value, the process also includes: Obtain the average value of all energy intensity values in the energy spectrum vector, and calculate the average energy value; Obtain the preset energy threshold factor in the operating system kernel; The peak energy ratio is obtained by calculating the ratio of the highest energy intensity value to the average energy value. Determine whether the peak energy ratio is greater than the energy threshold factor. If it is, confirm that the frequency component is a valid fault frequency component and extract the frequency corresponding to the valid fault frequency component as the center frequency of the main frequency band. If the peak energy ratio is not greater than the energy threshold factor, the center frequency of the main frequency band will be set to a null value or a preset normal frequency value.
[0069] The operating system kernel scheduler activated the "TASK_FFT_ANLYS" task with priority 30 when the UNIX timestamp 1761658201 arrived.
[0070] The task first calls the Fast Fourier Transform (FFT) function library integrated in the kernel based on the monitoring task activation instruction.
[0071] This function library reads the data blocks obtained from S1 and stored in shared memory, especially the sequence of sampling points within them. indivual value).
[0072] The FFT function processes this. indivual The FFT output is a sequence of values, from which the frequency components and their corresponding energy intensity values are obtained. The energy intensity is obtained by calculating the square of the modulus of each element in the complex vector. Since the input signal is a real number, its spectrum has conjugate symmetry, so only the first few elements are considered. points (i.e.) We will analyze the data from each point.
[0073] This energy spectrum vector For example, an array containing 10241 energy intensity values. .
[0074] Combined with sampling frequency (20480Hz) and the length of the sampling point sequence (20480), calculate frequency resolution .
[0075] .
[0076] Construct a frequency dimension representation, which includes: Frequency resolution: Hz.
[0077] Energy spectral vector: .
[0078] Frequency index list: .in .
[0079] Traversing the energy spectrum vector (from arrive ), to identify the frequency component with the highest energy intensity value.
[0080] In this embodiment, traversal It was found that in the index At that location, the energy intensity value is the highest. (Unit of energy, au)
[0081] In identification Next, perform the verification steps: Obtain the energy spectrum vector The average value of all 10241 energy intensity values is used to calculate the average energy value. .
[0082] .
[0083] In this embodiment, the calculation is obtained (au).
[0084] Obtain the preset energy threshold factor in the operating system kernel .
[0085] The setting is used to distinguish between the peak value of a real fault signal and the peak value of random noise.
[0086] For setting The following experiment was performed: Under the healthy operating condition of the wind turbine (such as state A in S2), 100 data blocks were collected and their energy spectra were calculated. The energy spectrum of each data block was calculated. The ratio was found to be lower than 99% in healthy samples. To ensure effective identification while allowing for a margin of error, Set as .
[0087] In this embodiment, .
[0088] Calculate the highest energy intensity value. With the average energy The ratio of the peak energy ratio is used to obtain the peak energy ratio. .
[0089] .
[0090] Determine the peak energy ratio (87.64) Is it greater than the energy threshold factor? (8.0).
[0091] The comparison result is "yes".
[0092] Therefore, confirm the index. The frequency component at that point is the effective fault frequency component.
[0093] Extract the effective fault frequency components from the frequency index list The corresponding frequency is used as the center frequency of the main frequency band. .
[0094] .
[0095] (like (au), (au).
[0096] .
[0097] judge The comparison result is "No".
[0098] Then the center frequency of the main frequency band Set to the preset normal frequency value 。 The final output center frequency of S3 is: Hz.
[0099] Please see Figure 1 and Figure 5 S4: Obtain the frequency-time slice mapping table in the operating system kernel, query the processor time slice length from the frequency-time slice mapping table based on the center frequency of the main frequency band, and adjust the processor time slice length to the upper limit of the execution time of the monitoring task.
[0100] The frequency-time slice mapping table includes the fault characteristic frequency range, the normal operating frequency range, and the corresponding time slice quota; The specific steps for S4 are as follows: Obtain the frequency-time slice mapping table in the operating system kernel; Based on the center frequency of the main frequency band, query the frequency-time slice mapping table to match the fault characteristic frequency range or normal operating frequency range to which the center frequency of the main frequency band belongs. Retrieve the time slice quota corresponding to the matching fault characteristic frequency range or normal operating frequency range, and determine the time slice quota as the processor time slice length; Adjust the processor time slice length to the upper limit of the execution time of the monitoring task; The specific process for adjusting the processor time slice length is as follows: Obtain the base time slice quota retrieved from the frequency-time slice mapping table. ; Obtain the time D-domain kurtosis value With ravine warning line ; The processor time slice length is calculated using the following formula. : ; in, The sensitivity coefficient for time slice adjustment is preset in the operating system kernel, which is used to adjust the degree to which the time domain kurtosis value exceeds the kurtosis warning line and its impact on the processor time slice length.
[0101] Obtain the frequency-time slice mapping table from the operating system kernel's default configuration (e.g., the / kernel / sched_map.dat file mapped to memory when the kernel starts).
[0102] This mapping table is based on the structural dynamics characteristics and known fault frequency features of the wind turbine (Vestas V112-3.0MW).
[0103] Table 2 Frequency-Time Slice Mapping Table As shown in Table 2, this table defines different frequency ranges and their corresponding basic time slice quotas. .
[0104] Based on the main frequency band center frequency obtained in S3 Hz, query the frequency-time slice mapping table.
[0105] Hz matching the frequency range of "50.1-1000.0" (intermediate frequency fault).
[0106] Retrieve the time slice quota corresponding to the matching frequency range , determined as ms.
[0107] This ms is used as the base value for the processor time slice length. Adjustments.
[0108] This adjustment process obtains ms, and obtain the time-domain kurtosis value from S1. Obtain the kurtosis warning line from S2. .
[0109] At the same time, obtain the preset time slice adjustment sensitivity coefficient in the operating system kernel. .
[0110] The setting is used to adjust the kurtosis. Beyond the warning line The magnitude of the increase in time slices.
[0111] For setting The following simulation tests were performed: Set the kurtosis value Reaching the ravine warning line At (6.0), the expected value is... exist Add to the basis .
[0112] .
[0113] Substitute into the formula: ; ; ; .
[0114] Therefore, setting .
[0115] Through formula Calculate processor time slice length .
[0116] The calculation logic of this formula is: Based on (determined by frequency type), calculate Compared to Excess ratio .like Not exceeding ,but If the value is 0, the time slice does not increase. Exceed Then the excess ratio is multiplied by the sensitivity coefficient. , as an amplification factor for time slices.
[0117] The advantage of the formula is that it achieves a dual dynamic adjustment. The base length of the time slice is determined by the "type" of the fault (via the center frequency of the main band). The final length is determined by the "severity" of the fault (via kurtosis). The values are weighted. This allows the system to differentiate between "critical" and "critical" faults. High value and Allocate the most CPU resources in the fault band.
[0118] Substitute the parameter values from this embodiment into the formula: ms , , ; ; calculate ; calculate ; ; ms.
[0119] This result indicates that, due to (3.0976) did not exceed (4.5), final processor time slice length That is, the basic time-based quota (150ms).
[0120] (If S1) S3 Hz: ms, , , ; calculate calculate ; ; ms.
[0121] In this case, the time slice is dynamically adjusted to ms. Finally, the processor time slice length calculated in this embodiment is... ms, set to the upper limit of the execution time of the "TASK_FFT_ANLYS" monitoring task on the next execution.
[0122] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the scope of protection defined by the claims of the present invention.
Claims
1. A wind turbine condition monitoring method based on edge computing, characterized in that, Includes the following steps: S1: Obtain the data block formed by the vibration acceleration signal of the wind turbine collected by the edge computing node, and calculate the time domain kurtosis value of the data block; S2: Obtain the preset kurtosis warning line, kurtosis alarm line and standard scheduling cycle in the operating system kernel, compare the time domain kurtosis value with the kurtosis warning line and the kurtosis alarm line, determine whether the scheduling mode of the monitoring task is the standard scheduling cycle, half of the standard scheduling cycle or immediate activation, and generate a monitoring task activation command. S3: Based on the monitoring task activation instruction, call the fast Fourier transform to process the data block, generate frequency components and corresponding energy intensity values, construct a frequency dimension representation, and identify the frequency of the frequency component with the highest energy intensity value from the frequency dimension representation as the center frequency of the main frequency band. S4: Obtain the frequency-time slice mapping table in the operating system kernel, query the processor time slice length retrieved from the frequency-time slice mapping table based on the main frequency band center frequency, and adjust the processor time slice length to the upper limit of the execution time of the monitoring task.
2. The wind turbine condition monitoring method based on edge computing according to claim 1, characterized in that, The data block includes a sampling point sequence, sampling frequency, and acquisition timestamp. The monitoring task activation instruction specifically includes task priority, scheduling timestamp, and target task identifier. The frequency dimension representation includes frequency resolution, energy spectrum vector, and frequency index list. The frequency-time slice mapping table includes fault characteristic frequency range, normal operating frequency range, and corresponding time slice quota.
3. The wind turbine condition monitoring method based on edge computing according to claim 1, characterized in that, The specific steps of S1 are as follows: The data block is formed by acquiring the vibration acceleration signal of the wind turbine collected by the edge computing node, and the time domain mean and time domain standard deviation of the sampling point sequence included in the data block are calculated. The fourth central moment of the sampling point sequence is calculated based on the time domain mean and the time domain standard deviation. The time-domain kurtosis value of the data block is obtained by calculating the ratio of the fourth-order central moment to the fourth power of the time-domain standard deviation.
4. The wind turbine condition monitoring method based on edge computing according to claim 1, characterized in that, The specific steps of S2 are as follows: Obtain the preset kurtosis warning threshold, kurtosis alarm threshold, and standard scheduling cycle in the operating system kernel; Compare the time domain kurtosis value with the kurtosis warning line and the kurtosis alarm line to determine whether the time domain kurtosis value is less than the kurtosis warning line. If so, determine that the scheduling mode is the standard scheduling cycle. Determine whether the time domain kurtosis value is greater than or equal to the kurtosis warning line and less than the kurtosis alarm line. If so, determine that the scheduling mode is half of the standard scheduling cycle. Determine whether the time-domain kurtosis value is greater than or equal to the kurtosis alarm threshold; if so, determine that the scheduling mode is to be activated immediately. Based on the determined scheduling method, a monitoring task activation instruction is generated.
5. The wind turbine condition monitoring method based on edge computing according to claim 2, characterized in that, The specific steps of S3 are as follows: Based on the monitoring task activation command, the fast Fourier transform is invoked to process the sampling point sequence in the data block to obtain the frequency components and the corresponding energy intensity values. The frequency resolution is calculated by combining the sampling frequency and the length of the sampling point sequence, and a frequency dimension representation including the frequency resolution, the energy spectrum vector, and the frequency index list is constructed. Traverse the energy spectrum vector to identify the frequency component with the highest energy intensity value; Extract the frequency corresponding to the frequency component in the frequency index list and use it as the center frequency of the main frequency band.
6. The wind turbine condition monitoring method based on edge computing according to claim 1, characterized in that, The specific steps of S4 are as follows: Obtain the frequency-time slice mapping table in the operating system kernel; Based on the center frequency of the main frequency band, query the frequency-time slice mapping table to match the fault characteristic frequency range or normal operating frequency range to which the center frequency of the main frequency band belongs; Retrieve the time slice quota corresponding to the matched fault characteristic frequency range or the normal operating frequency range, and determine the time slice quota as the processor time slice length; The processor time slice length is adjusted to the upper limit of the execution time of the monitoring task.
7. The wind turbine condition monitoring method based on edge computing according to claim 3, characterized in that, The time-domain kurtosis value Calculated using the following formula: ; in, The total number of sampling points in the sampling point sequence. The sampling point sequence is the first... Vibration acceleration values at each sampling point The time-domain mean of the sampling point sequence.
8. The wind turbine condition monitoring method based on edge computing according to claim 4, characterized in that, The process of generating the monitoring task activation instruction also includes dynamically determining the task priority based on the scheduling method; Obtain the preset standard priority, medium priority, and highest priority in the operating system kernel; When the scheduling mode is the standard scheduling period, the task priority is set to the standard priority; When the scheduling method is half of the standard scheduling cycle, the task priority is set to the medium priority. When the scheduling mode is immediate activation, the task priority is set to the highest priority.
9. The wind turbine condition monitoring method based on edge computing according to claim 5, characterized in that, After identifying the frequency component with the highest said energy intensity value, the method further includes: Obtain the average value of all energy intensity values in the energy spectrum vector, and calculate the average energy value; Obtain the preset energy threshold factor in the operating system kernel; Calculate the ratio of the highest energy intensity value to the average energy value to obtain the peak energy ratio; Determine whether the peak energy ratio is greater than the energy threshold factor. If it is greater, confirm that the frequency component is a valid fault frequency component and extract the frequency corresponding to the valid fault frequency component as the center frequency of the main frequency band. If the peak energy ratio is not greater than the energy threshold factor, then the center frequency of the main frequency band is set to a null value or a preset normal frequency value.
10. The wind turbine condition monitoring method based on edge computing according to claim 6, characterized in that, The process of adjusting the processor time slice length is as follows: Obtain the basic time slice quota retrieved from the frequency-time slice mapping table. ; Obtain the time-domain kurtosis value With the aforementioned kurtosis warning line ; The processor time slice length is calculated using the following formula. : ; in, A time slice adjustment sensitivity coefficient is preset in the operating system kernel to adjust the degree of influence of the time domain kurtosis value exceeding the kurtosis warning line on the processor time slice length.
