Wind power bolt torque real-time verification and calibration method and system empowered by edge computing

Abstract

The application provides an edge computing-enabled wind power bolt torque real-time verification and calibration method and system, and belongs to the technical field of wind power generation. The method comprises the following steps: obtaining a bolt friction coefficient and a temperature compensation coefficient of a strain sensor through a tightening process; synchronously collecting strain, temperature, ultrasonic wave flight time and unit operating condition parameters in operation; respectively calculating a strain method axial force and an ultrasonic wave method axial force; judging a zero point drift state of an intelligent sensor through double axial force cross verification and adjacent bolt spatial consistency analysis; setting an adaptive verification threshold according to an operating condition partition for dynamic verification; updating a zero point correction amount when drift is identified, and performing hierarchical calibration when axial force abnormality is identified. Through multi-source sensing fusion and edge computing, the application realizes online identification and dynamic calibration of sensor drift and real loosening, solves the problem that a fixed threshold cannot adapt to operating condition changes, and improves the long-term reliability and verification precision of a force measuring instrument.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation technology, specifically relating to a method and system for real-time verification and calibration of wind power bolt torque enabled by edge computing. Background Technology

[0002] As large and complex electromechanical systems exposed to extreme weather environments for extended periods, the structural integrity and operational reliability of wind turbines have always been a focus of industry attention. In the overall structure of wind turbines, bolted connections are widely used in key components such as tower flanges, hubs, blade roots, and the main frame, bearing the crucial responsibility of transferring loads and maintaining structural rigidity. The accuracy and stability of the bolt axial force directly determine whether the fasteners can effectively resist fatigue damage caused by random wind loads, alternating gravity loads, and tower vibrations.

[0003] Chinese utility model patent CN207569015U discloses a bolt loosening detection device. It uses a sleeve with positioning markings on the outside of the bolt and nut, and visually inspects the alignment of these markings to determine if there is a macroscopic shift in axial force. While this approach provided a direct and low-cost initial screening method for wind power operation and maintenance during a specific historical period, it is essentially a static and qualitative monitoring method. It heavily relies on the frequency of manual inspections and the experience of operators, and cannot achieve real-time digital extraction and automatic early warning of axial force data.

[0004] Chinese invention patent application CN118190224A discloses an axial torque monitoring system for offshore wind turbine bolts. This system uses a sensor network to perceive the mechanical state of the bolts in real time, providing data support for the scientific assessment of bolt loosening. Chinese invention patent application CN117968921A utilizes visual recognition technology to acquire images of the pointer dial and uses algorithms to calculate the pointer deflection angle to monitor pressure values, achieving non-contact monitoring. Chinese invention patent application CN121350527A further integrates environmental monitoring sensors and axial force measurement sensors, and uses augmented reality technology and a centralized computing model to correct axial force deviations during the installation phase.

[0005] The performance of existing technologies in practical industrial applications reveals that, as wind power generation evolves towards larger capacity, deeper and more intelligent applications, the existing monitoring architecture and processing mechanisms are gradually showing the following technical problems.

[0006] In the actual operating environment of wind farms, sensors are subjected to harsh conditions such as strong electromagnetic interference, severe vibration, and high salt spray corrosion for extended periods. Physical components inevitably experience zero-point drift or non-linear sensitivity decay. Existing technologies often treat the monitoring process as a simple signal reporting logic, lacking an effective real-time verification mechanism at the sensing end. When drift errors occur in the sensing hardware, the system cannot distinguish whether the actual axial force of the bolt has changed or whether the sensor itself has failed. This contradiction between inaccurate measurement and imprecise judgment directly weakens the credibility of the monitoring data.

[0007] The load borne by wind turbine bolts is a typical non-stationary random process, varying drastically with operating parameters such as wind speed, rotor speed, and pitch angle. Existing technologies use fixed thresholds for anomaly detection, which are prone to false alarms or missed alarms when operating conditions change. Especially under transient conditions such as gust impacts and start-stop transitions, fixed thresholds cannot distinguish between dynamic load fluctuations and actual bolt loosening, significantly reducing the practicality of the monitoring system.

[0008] Existing technologies all rely on a single type of sensor for measurement and lack independent cross-validation mechanisms. When a sensor malfunctions or drifts, the system cannot identify it autonomously and can only rely on periodic manual verification, resulting in high maintenance costs and slow response times. Summary of the Invention

[0009] The purpose of this invention is to provide a method and system for real-time verification and calibration of wind turbine bolt torque enabled by edge computing, which solves the technical problems in existing wind turbine bolt monitoring technologies, namely, the inability to distinguish between sensor drift and actual loosening, and the inability of fixed thresholds to adapt to changes in operating conditions.

[0010] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: This invention provides a method for real-time verification and calibration of wind turbine bolt torque enabled by edge computing, comprising the following steps: S1: Parameter calibration steps; During bolt installation or maintenance, the coefficient of friction of the bolt thread pair is obtained through a tightening-loosening process. coefficient of friction with the supporting surface The temperature compensation coefficient of the strain sensor was obtained through temperature cycling experiments. .

[0011] During the bolt tightening process, torque With axial force The relationship is determined by both the friction of the threaded pair and the friction of the supporting surface. Torque The physical composition can be broken down into three parts: the torque required to overcome the pitch angle, the torque required to overcome the friction of the thread pair, and the torque required to overcome the friction of the support surface. The above relationship is expressed as:

[0012] in, This represents the torque applied to the bolt, in units of... ; This indicates the bolt pitch, in units of... ; This represents the axial force of the bolt, expressed in units of... ; This represents the friction coefficient of the threaded pair, and is dimensionless. This indicates the bolt's pitch diameter, in units of... ; Indicates the helix angle of the thread, in units of ; This represents the coefficient of friction of the supporting surface, which is dimensionless. Represents the equivalent friction radius of the supporting surface, in units of .

[0013] The three terms on the right-hand side of the above equation correspond to the torque overcoming the pitch helix angle, the frictional torque of the thread pair, and the frictional torque of the bearing surface, respectively. This equation can be simplified to... ,in The torque coefficient is dimensionless. This is the nominal diameter of the bolt, in units of... The purpose of introducing the nominal diameter is to make the torque coefficient dimensionless, facilitating engineering applications. Torque coefficient The expression is:

[0014] Apply at least two different levels of torque during bolt installation or periodic calibration. and Simultaneously measure the corresponding rotation angle increment and and the axial force obtained through strain signals and Due to the torque coefficient within the elastic range Since it is a constant, we can obtain:

[0015] For separation and Introducing the proportional relationship of friction coefficient ,in This is a constant predetermined through tribological experiments. For steel-to-steel contact surfaces, the ratio of the friction coefficient of the threaded pair to the friction coefficient of the bearing surface was measured using a standard friction testing machine under the same surface treatment and lubrication conditions. Statistical results of the experimental data show... Typically, it is taken to be between 0.8 and 1.2. Substituting this into the torque coefficient expression, the solution is:

[0016]

[0017] Temperature compensation coefficient The principle behind obtaining it is as follows.

[0018] The main factors affecting the output of a strain gauge by temperature include the temperature coefficient of resistance of the sensing grid material and the difference in thermal expansion coefficients between the measured material and the sensing grid material. The overall temperature effect can be approximated as a first-order linear relationship.

[0019] in, It represents the thermal strain caused by temperature changes and is dimensionless. This represents the temperature compensation coefficient, in units of... ; This indicates the current temperature, in units of... ; Represents the reference temperature, in units of .

[0020] Place the bolt equipped with the strain sensor in the temperature control chamber and record at least three different temperature points under no external load. The sensor outputs strain value The temperature-strain curve was fitted using the least squares method:

[0021] in, Indicates the number of temperature points; Indicates the first Temperature values ​​at each temperature point, in units of ; Indicates the first The strain values ​​measured at each temperature point are dimensionless. This represents the arithmetic mean of the strain values ​​at each temperature point, and is dimensionless.

[0022] S2: Synchronous acquisition steps; During wind turbine operation, strain signals, temperature signals, ultrasonic time-of-flight signals, and turbine operating parameters of the bolts are synchronously collected at a preset sampling frequency. These operating parameters include wind speed, rotor speed, active power, and blade pitch angle.

[0023] To ensure time alignment of multi-source data, the IEEE 1588 precision time protocol or GPS synchronization pulses are used to synchronize the clocks of each sensor module. During each acquisition cycle, the edge computing node generates a synchronization trigger command, driving the strain sensor, temperature sensor, and ultrasonic transducer to sample simultaneously. This synchronization mechanism ensures that the physical quantities are strictly aligned on the time axis during subsequent multi-source data fusion, avoiding calculation errors introduced by sampling time deviations.

[0024] S3: Steps for calculating axial force; The first axial force is calculated based on the strain signal and temperature compensation coefficient. The second axial force is calculated based on the ultrasonic flight time. .

[0025] Calculation of the first axial force using the strain method The specific method is as follows: First, temperature compensation is performed on the strain signal, and the compensated strain value is:

[0026] in, This represents the strain value after temperature compensation, and is dimensionless. This represents the strain value directly measured by the strain sensor, and is dimensionless.

[0027] According to Hooke's Law, axial force With strain The relationship is:

[0028] in, This represents the axial force calculated using the strain method, in units of... ; This represents the elastic modulus of the bolt material, in units of... ; This represents the cross-sectional area of ​​the bolt under stress, in units of... .

[0029] Calculation of the second axial force using ultrasonic method The principle is as follows. The relationship between the propagation time of ultrasound inside the bolt and the bolt length is: ,in This indicates the round-trip time of the ultrasonic wave within the bolt, expressed in units of... ; Indicates the current length of the bolt, in units of... ; The velocity of sound in the bolt material is expressed in units of 1000 m / s. When a bolt is subjected to axial tensile force, the bolt length changes, and the sound velocity also changes slightly due to the acoustoelastic effect. Considering both geometric elongation and the acoustoelastic effect, the relationship between axial force and the rate of change of flight time can be expressed as:

[0030] in, This represents the axial force calculated using the ultrasonic method, with units of . ; Represents the acoustic elasticity calibration coefficient, in units of This was obtained through calibration experiments; This represents the time difference between flight under load and without load conditions, in units of... ; The initial flight time under no-load conditions is expressed in units of 1. .

[0031] Acoustoelastic calibration coefficient The method for obtaining the value is as follows: Under laboratory conditions, a known axial force is applied to bolts of the same batch. Measure the corresponding flight time changes Calculate the calibration coefficients:

[0032] To ensure calibration accuracy, measurements were taken at at least five different load levels, and linear regression fitting was used to obtain the results. value.

[0033] S4: Drift detection steps; By comparing the first axial force With the second axial force The deviation, and the analysis of the first axial force of adjacent bolts The spatial consistency of the change trend is used to determine the zero-point drift state of the strain sensor.

[0034] The principle of spatial consistency analysis is as follows: Bolts on the same flange face bear similar load spectra, and their axial force variation trends should exhibit spatial consistency. This is because when the flange face is subjected to external loads, the load distribution among the bolts follows a stiffness distribution law, and the stress state of adjacent bolts has inherent consistency. If the axial force variation of a single bolt is inconsistent with the statistical characteristics of adjacent bolts, then this variation is likely due to sensor malfunction rather than a true change in axial force.

[0035] Define bolt nodes At any moment Relative to the initial time Change in axial force:

[0036] in, Indicates bolt node At any moment Relative to the initial time The change in axial force, in units of ; Indicates bolt node At any moment Axial force, in units of ; Indicates bolt node At the initial moment Axial force, in units of .

[0037] For nodes Define its neighborhood set For nodes The set of bolt nodes (excluding nodes) on the flange surface with a topological distance less than or equal to a preset value. (Itself). The topological distance between adjacent nodes can be determined based on their geometric positions on the flange face, and is usually selected from the nodes. Two to four bolts that are directly adjacent to each other. Calculate the nodes. The ratio of the deviation between the change in axial force and the mean change in axial force of neighboring nodes:

[0038] in, Indicates bolt node At any moment The deviation ratio is dimensionless; Represents the neighborhood set The arithmetic mean of the changes in axial force at each node is given in units of 1000 m / s. ; Represents the neighborhood set The standard deviation of the axial force variation at each node, in units of When the axial force changes of all nodes in the neighborhood set are exactly equal, At this point, a direct judgment is made. If the value is infinity, a drift alarm will be triggered.

[0039] when When the threshold is exceeded, the node is determined. The strain sensor exhibits zero-point drift. The first preset threshold is set to 3 based on statistical distribution principles, corresponding to a 99.7% confidence level, i.e., when the node... A change in a value that deviates from the neighborhood mean by more than three standard deviations is considered a statistical anomaly.

[0040] At the same time, the first axial force With the second axial force Compare and calculate the relative deviation:

[0041] in, This represents the relative deviation between the axial force obtained by the strain method and the axial force obtained by the ultrasonic method; it is dimensionless. When the threshold is exceeded, it further confirms that the sensor is malfunctioning. The second preset threshold is determined based on the sensor's accuracy and engineering requirements, and is typically set to 5%. and The deviation is mainly due to This is caused by deviation from historical benchmarks, and If the deviation from the historical benchmark is relatively small (e.g., the deviation is less than the axial force change corresponding to the second preset threshold), it is determined to be strain sensor drift; otherwise, if It also deviates from the historical benchmark simultaneously, and the degree of deviation is the same as If the values ​​are similar, then it is determined to be a true change in axial force.

[0042] S5: Dynamic verification steps; Identify the current operating condition zone based on the unit's operating parameters, obtain the statistical distribution characteristics of historical axial force values ​​within the zone, set an adaptive verification threshold, and determine whether the bolt axial force status is abnormal.

[0043] The method for establishing operating condition zones is as follows: Collect operating condition parameter vectors acquired during historical normal operation. ,in Wind speed is expressed in units of 1. ; Indicates rotor speed, in units of ; Represents active power, in units of ; Represents the pitch angle, in units of .

[0044] For the initial stage of system operation where there is no historical data, the initial operating condition partition is established using design operating condition parameters or factory calibration data, and then updated online after sufficient operating data has been accumulated.

[0045] The working space is divided into K-means clustering algorithm. Each partition Number of partitions Determined by contour coefficients, typically 5 to 10.

[0046] In each partition Internal, statistical axial force mean and standard deviation To adapt to the slow evolution of bolt condition during long-term operation, and Online updates using the sliding window method:

[0047]

[0048] in, Indicates partition At any moment The average axial force, in units of ; Indicates partition At any moment The standard deviation of axial force, in units of ; This represents the preset learning rate, which is dimensionless and ranges from 0.01 to 0.1, controlling the weight of historical data and current data. Indicates time The measured axial force, in units of .

[0049] During real-time operation, based on current operating parameters Identify the partition The partition identification is based on minimizing the Euclidean distance between the current operating condition vector and the cluster centers of each partition:

[0050] in, Indicates partition The cluster center vector.

[0051] Set the adaptive verification threshold as follows:

[0052]

[0053] in, Indicates time The lower verification threshold, in units of ; Indicates time The upper verification threshold, in units of ; This represents a preset multiplier parameter, dimensionless, typically ranging from 3 to 5, corresponding to a confidence interval of 99.7% to 99.999%. When the measured axial force... If the value exceeds this dynamic threshold, it is considered abnormal.

[0054] S6: Calibration steps; When step S4 identifies zero-point drift, the zero-point drift correction is updated based on the multi-source data fusion results; when step S5 identifies an abnormal real axial force, an early warning or calibration command is issued.

[0055] Zero drift correction The update uses a weighted moving average, incorporating multiple criteria:

[0056] in, Indicates time The zero-point drift correction is dimensionless. , , The fusion weights are dimensionless and satisfy the following conditions: The weights can be dynamically adjusted based on the confidence level of each criterion; This represents the zero-point drift correction amount determined based on the downtime period; it is dimensionless. This represents the zero-point drift correction based on spatial consistency inference, and is dimensionless.

[0057] Drift judgment based on downtime The calculation method is as follows: when the unit is in a stopped state and the second axial force When the deviation from the historical benchmark value is less than the second preset threshold, it indicates that the actual load on the bolt has not changed significantly. In this case, the output change of the strain sensor can be attributed to zero-point drift.

[0058] in, This represents the zero-point drift correction amount determined based on the downtime period; it is dimensionless. This represents the reference strain value recorded during calibration; it is dimensionless.

[0059] Drift inference based on spatial consistency The calculation method is as follows: When bolt joint When the strain change differs significantly from the average strain change of neighboring bolt nodes, the inferred drift is:

[0060] in, This represents the zero-point drift correction based on spatial consistency inference, and is dimensionless. Indicates bolt node The current strain value is dimensionless; This represents the number of bolt nodes in the neighborhood set.

[0061] When a genuine axial force anomaly is identified, the system executes a tiered response based on the severity of the anomaly. The first level is software self-calibration: when the offset is within a preset small range, it is eliminated by adjusting the zero-point parameters of the verification algorithm. The second level is early warning calibration: when the offset exceeds a preset threshold, the system automatically generates a calibration command, guiding maintenance personnel to perform physical reinforcement using digital fastening tools. The third level is emergency shutdown: when the offset reaches a dangerous threshold, the system sends a shutdown request to the unit's main control system to prevent structural failure.

[0062] In addition, the present invention also provides a real-time verification and calibration system for wind power bolt torque enabled by edge computing, including a perception and identification layer, an edge processing layer and a collaborative management layer.

[0063] The sensing and identification layer is deployed at key connection points of the wind turbine and includes at least one bolt monitoring unit. Each bolt monitoring unit integrates a strain sensor, a temperature sensor, and an ultrasonic transducer to acquire the bolt's strain signal, temperature signal, and ultrasonic time-of-flight signal in real time. The strain sensor uses a full-bridge resistance strain gauge with a sensitivity of no less than [missing information - likely a specific value]. The temperature sensor uses a platinum resistance thermometer or a negative temperature coefficient thermistor, with an accuracy of no less than [a certain value]. The ultrasonic transducer uses piezoelectric ceramic material and operates in a frequency range of [frequency range missing]. to .

[0064] The edge processing layer is located inside the wind turbine tower or nacelle and is connected to the sensing and identification layer via an industrial bus to perform the above methods.

[0065] The edge processing layer includes a parameter calibration module, a signal processing module, an axial force calculation module, a drift discrimination module, a dynamic verification module, and a calibration execution module. The parameter calibration module is used to perform online identification of the friction coefficient and temperature compensation coefficient during bolt installation or maintenance. The signal processing module is used to filter, temperature compensate, and extract features from the acquired raw signals.

[0066] The axial force calculation module calculates the axial force based on strain and ultrasonic signals respectively. The drift discrimination module determines the sensor status through cross-validation and spatial consistency analysis. The dynamic calibration module sets adaptive thresholds based on operating condition zones and identifies axial force anomalies. The calibration execution module updates the zero-point correction or issues warning commands based on the drift discrimination results.

[0067] The collaborative management layer is deployed in a remote monitoring center and interacts with the edge processing layer via wireless communication links. It is used to store historical data, update global model parameters, and distribute verification policies. The collaborative management layer includes a global lifecycle database, a model evolution control unit, and a risk decision support system.

[0068] Compared with the prior art, the present invention has the following beneficial effects: Traditional strain gauge measurement methods directly respond to the elastic deformation of bolts, but cannot distinguish whether the deformation originates from preload decay or sensor zero-point drift. This invention introduces ultrasonic time-of-flight measurement as an independent physical observation dimension. The measurement benchmark for the ultrasonic method is the geometric length change of the bolt, a physical quantity unaffected by the strain gauge's bonding state or bridge imbalance. When a deviation occurs between the strain gauge measurement and the ultrasonic measurement, this deviation essentially reflects the offset of the strain sensor's reference zero point. Through cross-validation of two measurement methods with completely different physical principles, this invention achieves redundancy and cross-calibration of the signal source at the measurement principle level, fundamentally solving the inaccuracy problem.

[0069] This invention divides the working space into several partitions and establishes a statistical model of axial force distribution within each partition. The physical essence of this approach is to transform a non-stationary random process into a combination of multiple locally stationary processes, ensuring that the threshold setting conforms to the actual load distribution under each working condition. Compared to a fixed threshold, the working condition adaptive threshold is physically more consistent with the essential characteristics of bolt stress, thus improving the sensitivity of anomaly detection without increasing the false alarm rate.

[0070] This invention achieves online identification of sensor health status at the information fusion level. Spatial consistency analysis utilizes the mechanical continuity principle of flange connection structures. The load distribution of bolts on the same flange surface follows a stiffness distribution law, and the force changes of adjacent bolts have inherent consistency. When the trend of a single bolt deviates significantly from the statistical characteristics of its neighborhood, the deviation can only physically originate from sensor malfunction rather than actual load changes. This invention transforms physical laws into mathematical criteria, realizing online autonomous identification of sensor health status and providing a clear technical basis for system maintenance.

[0071] This invention utilizes tightening-loosening data during installation to identify the actual friction coefficient of each bolt online, reducing the uncertainty of model parameters to within individual differences and bringing the calculation accuracy of the physical model close to the theoretical limit of the measuring instrument. Dual-channel measurement using ultrasonic and strain methods creates observational redundancy, spatial consistency analysis creates logical redundancy, and self-calibration during downtime creates temporal redundancy. The integration of these three redundancy mechanisms allows the system to maintain basic functionality even in the event of a single sensor failure or partial data loss, improving the engineering reliability and environmental adaptability of the wind power bolt monitoring system. Attached Figure Description

[0072] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0073] Figure 1 This is an overall flowchart of the method described in this invention.

[0074] Figure 2 This is a flowchart of the zero-point drift identification and correction process of the strain method of the present invention.

[0075] Figure 3 This is a flowchart of the optimal estimation process for axial force through multi-source data fusion in this invention. Detailed Implementation

[0076] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0077] The following is in conjunction with the appendix Figures 1-3 The embodiments of the present invention will be described in detail below.

[0078] Example 1: This example provides a real-time verification and calibration system for wind turbine bolt torque enabled by edge computing. The system is constructed as a multi-level distributed architecture, which includes a perception and recognition layer, an edge processing layer, and a collaborative management layer from bottom to top.

[0079] The sensing and recognition layer is deployed at key connection points of the wind turbine.

[0080] In this embodiment, the flange connection at the top of the tower of a 5 MW offshore wind turbine is used as an example. A total of 120 M42 high-strength bolts are installed at the flange connection, and each bolt is equipped with a bolt monitoring unit. Each bolt monitoring unit integrates a strain sensor, a temperature sensor, and an ultrasonic transducer.

[0081] The strain sensor employs a full-bridge resistance strain gauge, fitted at a predetermined position on the bolt body. This strain gauge consists of four resistance strain gauges forming a Wheatstone bridge structure, effectively counteracting interference from non-axial stress. The strain sensor's sensitivity is... With a range covering 0 to 10,000 micro-strain, it is sufficient to cope with load fluctuations of wind turbines under extreme gust conditions.

[0082] The temperature sensor is a PT100 platinum resistance thermometer, mounted close to the bolt root. The PT100 platinum resistance thermometer... to Wide temperature range The high measurement accuracy provides the basic environmental parameters for subsequent temperature drift compensation. The temperature sensor and strain sensor are installed at the same point to ensure that the temperature measurement accurately reflects the temperature state at the location of the strain gauge.

[0083] The ultrasonic transducer uses piezoelectric ceramic material and operates at a frequency of [frequency missing]. It is fitted to the end of the bolt. The ultrasonic transducer acts as both an ultrasonic transmitter and receiver, measuring the round-trip time of the ultrasonic wave within the bolt using the pulse-echo method. The mounting surface of the ultrasonic transducer is precision-ground to ensure good acoustic coupling with the bolt end face.

[0084] All sensor signals are converted into digital signals via a local analog-to-digital converter and transmitted to the edge processing layer via shielded cable. To ensure interference immunity during signal transmission, the cable employs a double-shielded structure: the inner shield suppresses electric field interference, and the outer shield suppresses magnetic field interference. An active impedance matching circuit is configured at the input of the edge processing layer to perform impedance conversion on the weak differential signals output by the sensors at the source, reducing their output impedance to below 10 ohms.

[0085] The edge processing layer includes an edge computing gateway, housed in an electrical cabinet approximately 10 meters from the flange face inside the wind turbine tower. The edge computing gateway utilizes an industrial-grade processor with a built-in floating-point unit and operates over a wide temperature range. to The enclosure protection rating is no less than IP67. The gateway enclosure is made of die-cast aluminum alloy, which has good thermal conductivity and electromagnetic shielding performance. The edge computing gateway integrates a parameter calibration module, a signal processing module, an axial force calculation module, a drift discrimination module, a dynamic verification module, and a calibration execution module to execute the verification and calibration method of this invention.

[0086] The collaborative management layer, deployed in a remote monitoring center, comprises a global lifecycle database, a model evolution control unit, and a risk decision support system. It interacts with the edge processing layer via a 5G wireless network. The global lifecycle database stores the axial force evolution trajectory of each bolt node throughout its entire lifecycle since installation, providing data support for long-term fatigue analysis. The model evolution control unit employs an incremental learning architecture, periodically pushing updated verification weight coefficients to the edge processing layer based on regression analysis of massive amounts of historical data. The risk decision support system, based on verification anomaly signals uploaded by the edge processing layer, performs risk level determination and generates structured task messages, which are then sent to the maintenance personnel's terminals.

[0087] Example 2: In this example, the specific implementation process of online friction coefficient identification is described in detail. During the installation stage of the M42 high-strength bolts at the root of a 5 MW wind turbine blade, online friction coefficient identification is performed. The geometric parameters of the bolt are as follows: Nominal diameter pitch median diameter Equivalent friction radius of the support surface thread helix angle .

[0088] A digital electric torque wrench is used to apply torque to the bolts.

[0089] First apply the first torque level. At this time, the axial force is measured by a strain sensor. .

[0090] Continue applying the second torque level At this time, the axial force is measured by a strain sensor. .

[0091] Both sets of data were collected within the elastic deformation range of the bolt to ensure that the torque coefficient remains constant.

[0092] Calculate the dimensionless torque coefficient :

[0093] The ratio of friction coefficients between steel-steel contact surfaces, determined beforehand by tribological experiments. Substitute into the formula to calculate the friction coefficient of the threaded pair. :

[0094] First, calculate each item: ; ; ; ; denominator ; ; Calculate the friction coefficient of the support surface : ; The coefficient of friction obtained by identification and The parameters are written to the local storage module of the edge computing gateway as individual parameters for the bolt, and used for axial force verification during subsequent operation.

[0095] Example 3: This example details the specific implementation process of temperature compensation coefficient calibration. Specifically, it is as follows: An M42 bolt equipped with a strain sensor was placed in a temperature control chamber, and a temperature cycling experiment was conducted under no external load. The temperature control range of the chamber was [temperature range to be specified]. to Temperature control accuracy is During the experiment, with As a reference temperature .

[0096] Record the output values ​​of the strain sensor at different temperature points.

[0097] The experimental data are as follows: temperature At that time, strain Micro-strain; temperature At that time, strain Micro-strain; temperature At that time, strain Micro-strain; temperature At that time, strain Micro-strain; temperature At that time, strain Micro-strain.

[0098] Calculate the temperature difference between each temperature point and the reference temperature. and the corresponding strain values The temperature-strain curve was fitted using the least squares method:

[0099] in The arithmetic mean of the strain values ​​at each temperature point: Micro-strain; Calculate the numerator:

[0100] Calculate the denominator:

[0101] Temperature compensation coefficient Micro-strain per degree Celsius; This temperature compensation coefficient Write the data to the local storage module of the edge computing gateway for temperature compensation of strain signals during subsequent operation.

[0102] Example 4: This example mainly describes in detail the ultrasonic acoustoelastic calibration coefficient. The calibration process.

[0103] Under laboratory conditions, M42 high-strength bolts of the same batch and specification as the bolts used on site were taken as calibration samples. The bolts were mounted on a tensile testing machine, which could apply precisely controlled axial force, with a measurement accuracy of [insert accuracy here]. An ultrasonic transducer is attached to the end of the bolt, and the round-trip time of the ultrasonic wave within the bolt is measured using a pulse transmitter and receiver.

[0104] The calibration process is as follows: First, measure the initial flight time of the bolt under no-load conditions. Gradually increase the load on the tensile testing machine, increasing it by a certain amount each time. Record one axial force and corresponding flight time Continue until the bolt's rated load is reached. Repeat the measurement three times for each load level and take the average value.

[0105] The calibration data is recorded in Table 1 below: Table 1:

[0106] According to the formula It can be known that and The relationship is linear, and the slope is... Perform linear regression (forced to cross zero) on the above data to obtain the slope. .

[0107] The acoustic elastic calibration coefficient Write the data to the local storage module of the edge computing gateway for subsequent calculation of the ultrasonic normal axial force during operation.

[0108] Example 5: This example uses a single M42 bolt at the flange connection of a 5 MW wind turbine tower as an example to describe in detail the entire process of online monitoring and verification.

[0109] The wind turbine operates under rated conditions with a wind speed of The rotor speed is The active power is The pitch angle is Edge computing gateways The sampling frequency is used to synchronously acquire strain signals, temperature signals, and ultrasonic time-of-flight signals.

[0110] Synchronous acquisition: Under the synchronization mechanism of the IEEE 1588 precision time protocol, the strain sensor, temperature sensor, and ultrasonic transducer complete sampling under the same trigger pulse. Each acquisition cycle is... This ensures strict alignment of multidimensional data on the timeline.

[0111] Axial force calculation: The strain sensor measures the raw strain value. Micro-strain. The temperature sensor measures the real-time temperature. Reference temperature Temperature compensation coefficient Calculate the strain value after temperature compensation:

[0112] Bolt elastic modulus Cross-sectional area under stress Calculate the first axial force. :

[0113] The time of flight under load is measured by an ultrasonic transducer. Initial flight time without payload Flight time variation Acoustoelastic calibration coefficient Calculate the second axial force. :

[0114] Drift detection: Calculate the bolt node The change in axial force at the initial moment. Axial force This value is the normal preload force calibrated and recorded using the torque method during bolt installation, at the current moment. Axial force Change .

[0115] Obtain the axial force data of adjacent bolts on the flange face. (Select the neighboring region.) Includes nodes There are 4 bolts (excluding the bolt joints) at adjacent bolt nodes on the flange face. (Itself). The changes in axial force at each node are as follows: , , , .

[0116] Calculate the mean value of the change in axial force in the neighborhood:

[0117] Calculate the standard deviation of the variation in axial force in the neighborhood:

[0118]

[0119] Calculate the deviation ratio :

[0120] The first preset threshold is set to 3. A value much greater than 3 indicates that the axial force variation of this bolt differs statistically from that of neighboring bolts, consistent with the characteristics of sensor drift.

[0121] Further comparison of the first axial force With the second axial force Calculate the relative deviation:

[0122] The second preset threshold is 5%. It far exceeds 5%. Deviation from historical benchmark ( ) ,and Deviation from historical benchmark ( )only , The relatively small deviation indicates that the strain sensor has zero-point drift while the actual axial force changes little.

[0123] Based on the above analysis, it is determined that the strain sensor of the bolt exhibits zero-point drift.

[0124] Dynamic verification is as follows: Identify current operating parameters: wind speed Rotor speed Active power Pitch angle .

[0125] Based on historical data, K-means clustering was used to partition the work conditions. The current work condition belongs to partition [number missing]. .

[0126] This partition The historical axial force statistics are as follows: mean Standard deviation .Pick Set the adaptive verification threshold as follows:

[0127]

[0128] Current measured axial force Far below the lower threshold The axial force was determined to be abnormal. However, according to the drift discrimination result of step S4, the abnormality has been identified as caused by sensor drift, rather than a decrease in actual axial force.

[0129] Calibration is as follows: Since the sensor has been identified as zero-point drift, the system performs zero-point drift calibration.

[0130] Based on the drift judgment during the shutdown period: If the unit is currently in operation and does not meet the shutdown conditions, this item will not be included in the integration for the time being.

[0131] Drift inference based on spatial consistency: calculation : Micro-strain, the average strain of the neighboring bolts is calculated back from the axial force of the neighboring area to be approximately Micro-strain, Micro-strain.

[0132] Update the zero-point drift correction. (Take...) , Data is unavailable during downtime. The last drift correction amount Micro-strain:

[0133] The correction amount is written to local storage, and subsequent strain measurements will be subtracted from this correction amount for compensation. Simultaneously, the system generates a secondary warning tag indicating a decline in sensor health, reminding maintenance personnel to check the sensor's electrical connections during the next inspection.

[0134] Example 6: This example verifies the zero-point drift calibration capability of the present invention during unit shutdown. Zero-point drift calibration was performed during the maintenance shutdown of a 5 MW wind turbine unit.

[0135] The unit is in a shutdown state, and the wind speed is lower than the cut-in wind speed. The rotor speed is zero. At this time, the second axial force... If the deviation from the historical benchmark value is less than 5%, it indicates that the actual load on the bolt has not changed significantly.

[0136] The strain sensor measures the raw strain value. Micro-strain. The reference strain value recorded during calibration. Micro-strain. Real-time temperature. reference temperature Temperature compensation coefficient Calculate the drift correction based on the downtime period:

[0137] Update the zero-point drift correction. (Take...) , , The last drift correction amount , :

[0138] After calibration, remeasure the strain output in the shutdown state and correct the strain value. The micro-strain, close to zero, verifies the effectiveness of the calibration.

[0139] Example 7: During the operation of a 5 MW wind turbine, the axial force of a bolt was monitored. from Descending to The second axial force measured by ultrasonic method from Descending to relative deviation It is less than 5% of the second preset threshold. Calculate the deviation ratio. If the value is less than the first preset threshold of 3, it indicates that the trend of change of the bolt is consistent with that of the neighboring bolts.

[0140] Based on the above information, it was determined that the issue was a genuine decrease in axial force rather than sensor drift. The system triggered a second-level early warning calibration, generating a calibration command and pushing it to the maintenance personnel's terminal. Upon arrival at the site, the maintenance personnel used digital fastening tools to reinforce the bolts, restoring the axial force to its rated value.

[0141] Example 8: Based on Example 5, this example further provides an optimal axial force estimation method based on multi-source data fusion to solve the confidence difference between strain method measurement and ultrasonic method measurement under different working conditions, and to achieve optimal fusion estimation of axial force.

[0142] The strain method has the advantages of fast response and high sampling rate, but it is greatly affected by temperature drift and zero drift; the ultrasonic method has the advantages of good stability and no influence from zero drift, but the sampling rate is low and it is affected by the quality of the echo signal.

[0143] This embodiment establishes a dynamic confidence evaluation model and uses a Kalman filter framework to achieve adaptive fusion of two measurement methods.

[0144] The optimal estimation method for axial force based on multi-source data fusion includes the following steps: Step S5-1: Confidence assessment of strain method measurement; The confidence level of strain gauge measurements is assessed by comprehensively considering the following factors: zero-point drift state, temperature compensation error, and historical consistency.

[0145] Zero drift confidence factor The calculation formula is:

[0146] in, This represents the zero-point drift confidence factor, which is dimensionless. This represents the current zero-point drift correction amount, which is dimensionless. This represents the maximum allowable value for zero-point drift correction, which is determined based on the sensor's range and is set to 500 micro-strain.

[0147] Temperature Compensation Confidence Factor The calculation formula is:

[0148] in, This represents the temperature compensation confidence factor, which is dimensionless. This indicates the current temperature, in units of... ; Represents the reference temperature, in units of ; Indicates the effective range of temperature compensation, taking... .

[0149] Historical consistency confidence factor The calculation formula is:

[0150] in, This represents the confidence factor for historical consistency, and is dimensionless. This represents the current change in axial force using the strain gauge method, in units of... ; This represents the moving average of the historical axial force variation, in units of... ; The standard deviation of the historical axial force variation is expressed in units of 1000 ppm. .

[0151] Overall confidence level of strain method The calculation formula is:

[0152] in, The overall confidence level of the strain method is dimensionless. , , The values ​​represent the weight coefficients of each factor, which are dimensionless and are taken as 0.4, 0.3, and 0.3 respectively.

[0153] Step S5-2: Confidence assessment of ultrasonic measurement; The confidence level of the ultrasonic method is measured by the echo signal quality index. Evaluate consistency with time.

[0154] echo quality confidence factor Equal to echo signal quality index The range of values ​​is .

[0155] Time consistency confidence factor The calculation formula is:

[0156] in, This represents the confidence factor for time consistency, and is dimensionless. This indicates the change in current flight time, in units of... ; This represents the moving average of the historical flight time variation, in units of... ; The standard deviation of the variation in historical flight time is expressed in units of 1000 m / s. .

[0157] Overall confidence level of ultrasonic method The calculation formula is:

[0158] in, This represents the overall confidence level of the ultrasonic method and is dimensionless. , The values ​​represent the weight coefficients of each factor, which are dimensionless and are taken as 0.6 and 0.4 respectively.

[0159] Step S5-3: Kalman filter fusion estimation; A Kalman filter framework is used to perform optimal fusion estimation of the axial force. A state-space model is established: Equations of state:

[0160]

[0161] in, Indicates time The actual axial force, in units of ; Represents the rate of change of axial force, in units of ; Indicates the sampling time interval, in units of ; The process noise representing axial force follows a mean of 0 and a variance of . The normal distribution; The process noise representing the rate of change of axial force follows a mean of 0 and a variance of . It follows a normal distribution.

[0162] Observation equation:

[0163]

[0164] in, Indicating the strain method at time The axial force measurement value, in units of ; Indicates the ultrasonic method at time The axial force measurement value, in units of ; The observation noise of the strain gauge method is represented by a mean of 0 and a variance of . The normal distribution; The observation noise of the ultrasonic method is represented by a mean of 0 and a variance of . It follows a normal distribution.

[0165] The variance of the observation noise is dynamically adjusted based on the overall confidence level.

[0166]

[0167] in, Indicates time The observation noise variance of the strain method, in units of ; This represents the initial observation noise variance of the strain gauge method, obtained through calibration experiments, with units of [unit missing]. ; Indicates time The overall confidence level of the strain method is dimensionless. Indicates time The observation noise variance of the ultrasonic method, in units of ; This represents the initial observation noise variance of the ultrasonic method, obtained through calibration experiments, and is expressed in units of... ; Indicates time The overall confidence level of the ultrasonic method is dimensionless.

[0168] The Kalman filter update process includes two stages: prediction and correction. Prediction phase:

[0169]

[0170] in, Indicates time Prior estimate of axial force, in units of ; Indicates time The posterior estimate of the axial force, in units of ; Indicates time The posterior estimate of the rate of change of axial force, in units of ; This represents the prior estimate error covariance, in units of... ; This represents the covariance of the posterior estimation error, in units of... ; This represents the process noise covariance matrix.

[0171] Correction phase:

[0172]

[0173]

[0174] in, Represents the Kalman gain matrix; Represents the observation matrix; Represents the observation noise covariance matrix; Indicates time The posterior estimate of the axial force, i.e., the best estimate after fusion, in units of... ; This represents the observation vector, which includes measurements from both the strain method and the ultrasonic method. Represents the identity matrix.

[0175] Step S5-4: Output the confidence interval of the fusion result; Based on the fused estimation error covariance Calculate the 95% confidence interval for the optimal estimate of the axial force:

[0176] in, This represents the optimal estimate of the axial force and its confidence interval, in units of... .

[0177] Through the multi-source data fusion method in this embodiment, the overall error of axial force measurement is reduced from that of a single method. Reduce to It also provides measurement uncertainty assessment, providing a more reliable basis for subsequent axial force anomaly determination.

[0178] In some preferred embodiments, in addition to the tightening-loosening method described above, the following alternatives can be used for online identification of the coefficient of friction.

[0179] This method utilizes the difference in sensitivity of longitudinal and transverse wave propagation velocities in bolts to axial stress, simultaneously measuring the flight times of both waves, and then calculating the axial stress and coefficient of friction by the ratio of the changes in the two wave velocities. This method does not require the application of different levels of torque and can complete identification during a single tightening process, making it suitable for automated assembly scenarios.

[0180] This method utilizes slope analysis based on the torque-rotation angle curve. During bolt tightening, the complete torque-rotation angle curve is recorded. Within the elastic segment, the slope of the curve exhibits a definite functional relationship with the coefficient of friction. By fitting the slope of the elastic segment and combining it with known thread geometry parameters, the coefficient of friction can be inversely calculated. This method is applicable to electric tightening tools with rotation angle measurement capabilities.

[0181] For calibrating the temperature compensation coefficient, in addition to the temperature chamber test method, the following alternative methods can be used: Online identification method based on historical data. During the long-term operation of the wind turbine, a large amount of strain and temperature data is collected during shutdown periods. During shutdown periods, the bolts do not experience actual load changes; strain changes are entirely caused by temperature. The temperature compensation coefficient is updated online through regression analysis. This method requires no additional experiments and can adapt to long-term changes such as material aging.

[0182] In addition to clustering algorithms, the following alternative approach can be used to establish operating condition zones: a physical model-based operating condition partitioning method. Based on the structural dynamics characteristics of the wind turbine, operating conditions are divided into physical states such as startup, rated operation, overload operation, and shutdown. An axial force statistical distribution is then established within each physical state. This method offers stronger physical interpretability and does not require a large amount of historical data.

[0183] The method and system provided by this invention can be widely applied to online monitoring of bolt axial forces in key connection parts of wind turbine generators, such as tower flanges, hubs, and blade roots. By using edge computing technology, complex verification computation is decentralized to the local unit, reducing data transmission latency and improving the system's response speed to sudden axial force anomalies. Through multi-source sensor fusion and dynamic calibration algorithms, the problem of environmental interference affecting accuracy is solved, providing a reliable technical means for structural health monitoring of wind turbine generators. The implementation of this invention will improve the intelligence level of wind farm operation and maintenance, reduce the risk of major structural failure accidents, and has good prospects for industrial application.

[0184] Example 9: Based on Example 5, this example further provides a dynamic temperature compensation method based on Physics-Informed Neural Network (PINN) to solve the problem that a fixed temperature compensation coefficient cannot adapt to sensor aging, adhesive layer creep and local thermal gradient changes.

[0185] Step S1-1: Construct a physical information neural network model; Deploy a lightweight physical information neural network in an edge computing gateway. The network structure is as follows: Input layer: 5 neurons, corresponding to the temperature values ​​of 5 consecutive sampling points within a time window. to (Sampling interval) ), strain value after compensation at the previous moment And normalized unit operating status flags (0 indicates shutdown, 1 indicates operation).

[0186] Hidden layers: 2 layers, 16 neurons per layer, using the Swish activation function to maintain smooth and differentiable properties.

[0187] Output layer: 2 neurons, each outputting the dynamic temperature compensation coefficient. (unit: ) and the predicted strain value at the current moment (unit: ).

[0188] Step S1-2: Embed the physical constraint loss term; The physical constraints are divided into three parts: (1) Physical constraints of thermal strain:

[0189] in The physical constraint loss term is defined as:

[0190] (2) Physical constraints on elastic deformation (downtime): During the unit downtime (wind speed < 3 m / s, rotor speed = 0), the bolts experience no external load changes, and strain changes are entirely caused by temperature. Therefore, the data collected during the downtime should meet the following requirements:

[0191] The physical constraint loss term is:

[0192] in This represents the number of samples during the downtime period.

[0193] (3) Smoothness constraint: To avoid abrupt changes in the compensation coefficients, a time smoothing loss is added:

[0194] The total physical constraint loss is:

[0195] in , , .

[0196] Steps S1-3: Online training and updates; The edge computing gateway incrementally trains the PINN model every 24 hours using historical data from the past 7 days (including runtime and downtime periods). The optimizer is Adam, with an initial learning rate of 0.001, a decay rate of 0.95 per epoch, and 50 training epochs. The training data volume is approximately 600,000 sets (7 days × 24 hours × 3600 seconds × 1Hz sampling).

[0197] The composite loss function is:

[0198] in For network output Compared with measured strain The mean square error.

[0199] After training is complete, the model parameters are updated to the edge processing layer for real-time inference.

[0200] Steps S1-4: Real-time dynamic compensation; In real-time operation, the input temperature sequence and the strain after compensation at the previous moment are used for each sampling period (1Hz), and the network outputs the current dynamic temperature compensation coefficient through forward inference. and predicting strain .

[0201] The temperature compensation correction in the first axial force calculation formula is as follows:

[0202] in This is the zero-point drift correction amount in step S6.

[0203] Step S1-5: Sensor health prediction; Long-term recording of dynamic temperature compensation coefficient The time series. When The drift exceeded the initial calibration value for 30 consecutive days. When the drift reaches 20%, the system generates a "sensor aging warning"; when the drift exceeds 50%, it generates a "sensor replacement suggestion".

[0204] Data from six months of continuous operation on the flange bolts of a 5 MW wind turbine tower shows: Initial calibration temperature compensation coefficient ; By the third month of operation, the PINN dynamic compensation coefficient had stabilized at ; After six months of operation, the PINN dynamic compensation coefficient drifted to The relative change was 12.9%, triggering a "sensor aging warning".

[0205] Under the same operating conditions (wind speed 12m / s, temperature 35℃), the axial force measurement accuracy and drift detection capability of fixed temperature compensation (Example 5) and dynamic PINN compensation (this example) are compared, as shown in Table 2.

[0206] Table 2:

[0207] Data source: An M42 bolt test bench was set up in a laboratory environment, and temperature cycling (-20℃ to 60℃) and axial force cycling (0 to 300 kN) were applied. The fixed compensation algorithm and the PINN dynamic compensation algorithm were run respectively, and data were collected for 30 consecutive days.

[0208] This embodiment addresses the technical challenge of offline temperature compensation calibration failing to adapt to long-term operation. In Embodiment 5, the temperature compensation coefficient is calibrated once during installation, but in actual operation, sensor adhesive aging, surface corrosion, and material creep alter thermal response characteristics, causing the fixed compensation to gradually fail. This embodiment introduces physical constraints on heat conduction and self-monitoring signals during downtime through a physical information neural network, enabling the temperature compensation coefficient to adaptively evolve with the sensor's state, maintaining high-precision compensation and fundamentally eliminating the problem of temperature compensation errors accumulating over time. Furthermore, through long-term trend analysis of the dynamic temperature compensation coefficient, this embodiment can predict sensor aging trends 45 days in advance, transforming passive maintenance into predictive maintenance. This is particularly important for deep-sea wind farms, significantly reducing unplanned downtime and inspection costs.

[0209] This embodiment introduces physical information neural networks into wind turbine bolt monitoring for the first time, embedding heat conduction equations and elasticity laws as prior knowledge into the network training, enabling the model to converge with a small amount of data and possess extrapolation capabilities.

[0210] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0211] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time verification and calibration of wind turbine bolt torque powered by edge computing, characterized in that, Includes the following steps: S1: During the bolt installation or maintenance phase, the friction coefficient of the bolt thread pair and the friction coefficient of the support surface are obtained through the tightening process, and the temperature compensation coefficient of the strain sensor is obtained through temperature cycling experiments. S2: During the operation of the wind turbine, the strain signal, temperature signal, ultrasonic time-of-flight signal and turbine operating parameters of the bolt are collected synchronously at a preset sampling frequency. S3: Calculate the first axial force based on the strain signal and the temperature compensation coefficient, and calculate the second axial force based on the ultrasonic flight time; S4: By comparing the deviation between the first axial force and the second axial force, and analyzing the spatial consistency of the changing trend of the first axial force of adjacent bolts, the zero-point drift state of the strain sensor is determined. S5: Identify the current operating condition zone based on the unit operating condition parameters, obtain the statistical distribution characteristics of historical axial force values ​​in the zone, set an adaptive verification threshold, and determine whether the bolt axial force status is abnormal. S6: When step S4 identifies zero-point drift, update the zero-point drift correction amount based on the multi-source data fusion result; when step S5 identifies axial force anomaly, issue a warning or calibration command.

2. The edge computing-enabled real-time verification and calibration method for wind turbine bolt torque according to claim 1, characterized in that, The methods for obtaining the friction coefficient of the threaded pair and the friction coefficient of the supporting surface in step S1 include: Apply at least two different levels of torque, simultaneously measure the corresponding axial force, and calculate the torque coefficient; By introducing a friction coefficient proportional relationship, and substituting this relationship into the torque coefficient expression, the friction coefficient of the threaded pair and the friction coefficient of the supporting surface can be solved. The torque coefficient is calculated based on the difference between the torque values ​​at the two different levels, the nominal diameter of the bolt, and the difference in the corresponding axial force.

3. The edge computing-enabled real-time verification and calibration method for wind turbine bolt torque according to claim 1, characterized in that, The calculation method for the second axial force in step S3 includes: The second axial force is calculated based on the product of the acoustoelastic calibration coefficient and the rate of change of ultrasonic flight time. The acoustoelastic calibration coefficients are obtained through laboratory calibration, and the ultrasonic flight time variation rate is the ratio of the flight time difference between the loaded and unloaded states to the initial flight time in the unloaded state.

4. The edge computing-enabled real-time verification and calibration method for wind turbine bolt torque according to claim 1, characterized in that, The spatial consistency analysis method described in step S4 includes: Define the change in axial force at the bolt joint at time t relative to the initial time. Define a neighborhood set for each bolt node, wherein the neighborhood set is the set of adjacent bolt nodes whose topological distance to the bolt node is less than or equal to a preset value; Calculate the deviation ratio between the axial force variation of the bolt node and the mean axial force variation of neighboring nodes. The deviation ratio is equal to the absolute value of the difference between the axial force variation of the bolt node and the mean axial force variation of neighboring nodes divided by the standard deviation of the axial force variation of neighboring nodes. When the deviation ratio exceeds the first preset threshold, it is determined that the strain sensor of the bolt node has zero-point drift.

5. The edge computing-enabled real-time verification and calibration method for wind turbine bolt torque according to claim 1, characterized in that, The method for comparing the deviation between the first axial force and the second axial force in step S4 includes: Calculate the relative deviation between the first axial force and the second axial force. The relative deviation is equal to the absolute value of the difference between the first axial force and the second axial force divided by the maximum value of the two and then multiplied by 100%. When the relative deviation exceeds the second preset threshold and the degree to which the first axial force deviates from the historical reference is greater than the degree to which the second axial force deviates from the historical reference, it is determined to be strain sensor drift; When the relative deviation exceeds the second preset threshold and the second axial force deviates synchronously from the historical benchmark, it is determined to be a true change in axial force.

6. The edge computing-enabled real-time verification and calibration method for wind turbine bolt torque according to claim 1, characterized in that, The method for establishing the working condition partition in step S5 includes: Collect operating condition parameter vectors collected during historical normal operation, including wind speed, rotor speed, active power, and blade pitch angle; Clustering algorithms are used to divide the working space into several partitions, and the number of partitions is determined by the profile coefficient. Within each partition, the mean and standard deviation of the axial force are calculated, and the mean and standard deviation are updated online using the sliding window method.

7. The edge computing-enabled real-time verification and calibration method for wind turbine bolt torque according to claim 1, characterized in that, The method for setting the adaptive verification threshold in step S5 includes: The partition is identified based on the current operating condition parameters. The partition identification criterion is to minimize the Euclidean distance between the current operating condition vector and the cluster centers of each partition. An adaptive verification threshold is set, which includes a lower verification threshold and an upper verification threshold. The lower verification threshold is equal to the average axial force of the current partition minus the product of a preset multiple parameter and the standard deviation of the axial force of the current partition. The upper verification threshold is equal to the average axial force of the current partition plus the product of a preset multiple parameter and the standard deviation of the axial force of the current partition. When the measured axial force exceeds the lower verification threshold or the upper verification threshold, it is determined to be abnormal.

8. The edge computing-enabled real-time verification and calibration method for wind turbine bolt torque according to claim 1, characterized in that, Step S5 also includes a method for optimal estimation of axial force through multi-source data fusion, specifically including: The overall confidence level of the strain method measurement is evaluated, and the overall confidence level is calculated based on a weighted average of the zero-point drift confidence factor, the temperature compensation confidence factor, and the history consistency confidence factor. The overall confidence level of the ultrasonic measurement is evaluated, and the overall confidence level is calculated based on a weighted average of the echo quality confidence factor and the time consistency confidence factor. A Kalman filter framework is used for optimal estimation of axial force, wherein the observation noise variance is dynamically adjusted based on the combined confidence level of the strain method measurement and the combined confidence level of the ultrasonic method measurement. The optimal estimate of the axial force is obtained by Kalman filtering.

9. The edge computing-enabled real-time verification and calibration method for wind turbine bolt torque according to claim 8, characterized in that, The zero-point drift confidence factor is calculated based on the ratio of the current zero-point drift correction to the maximum allowable drift. The temperature compensation confidence factor is calculated based on the deviation between the current temperature and the reference temperature, as well as the effective range of temperature compensation. The historical consistency confidence factor is calculated based on the deviation between the current strain gauge axial force variation and the historical axial force variation moving average, as well as the standard deviation of the historical axial force variation. The echo quality confidence factor is equal to the echo signal quality index. The time consistency confidence factor is calculated based on the deviation between the current flight time variation and the historical flight time variation moving average, as well as the standard deviation of the historical flight time variation.

10. A wind turbine bolt torque real-time verification and calibration system powered by edge computing, characterized in that, include: The sensing and identification layer is deployed at the key connection parts of the wind turbine and includes at least one bolt monitoring unit. Each bolt monitoring unit integrates a strain sensor, a temperature sensor and an ultrasonic transducer to collect the bolt's strain signal, temperature signal and ultrasonic time-of-flight signal in real time. An edge processing layer, located inside the wind turbine tower or nacelle, is connected to the sensing and identification layer via an industrial bus, and is used to execute the edge computing-enabled wind turbine bolt torque real-time verification and calibration method as described in any one of claims 1 to 9. The collaborative management layer is deployed in a remote monitoring center and interacts with the edge processing layer via wireless communication links. It is used to store historical data, update global model parameters, and issue verification policies.

Images