Millimeter-wave-based wind turbine blade measurement method

Abstract

This invention relates to the field of wind power equipment monitoring technology and discloses a method for measuring wind turbine blades based on millimeter waves. The method initializes a millimeter-wave radar system and sets up a multi-band transmission mode, transmitting millimeter-wave signals to the wind turbine blades and receiving reflected signals to obtain a digital signal stream. The digital signal stream is then adaptively filtered and phase-calibrated to generate enhanced signal data. This enhanced signal data is analyzed to extract the blade surface profile and internal structural features, constructing a blade feature dataset. Based on this dataset, blade size parameters and material property parameters are calculated to form a blade parameter set. The transmission power and receiving sensitivity of the millimeter-wave radar are dynamically adjusted to achieve adaptive optimization of the measurement process. The radar is controlled to continuously scan the blades according to the optimized parameters, acquiring time-series dynamic measurement data. The dynamic data is processed to detect abnormal blade vibrations and structural changes, generating blade condition measurement results.

Description

Technical Field

[0001] This invention relates to the field of wind power equipment monitoring technology, specifically a method for measuring wind turbine blades based on millimeter waves. Background Technology

[0002] Current wind turbine blade inspection primarily relies on manual visual inspection or fixed-point measurement using a single sensor. Existing technologies for monitoring blade condition largely depend on periodic shutdowns for inspection, failing to achieve real-time measurement during operation. Measurement methods are limited, typically only acquiring surface information and failing to detect internal structural defects. Data acquisition is heavily influenced by the environment; wind load vibration and weather changes increase measurement errors. Signal processing is simplistic, failing to effectively separate valid signals from environmental noise. Feature extraction dimensions are limited, failing to simultaneously acquire surface morphology and internal structural features. Parameter settings are fixed, and the operating parameters of the measuring equipment cannot be dynamically adjusted according to blade characteristics. Anomaly detection is delayed, often only discovered after damage is apparent, lacking early warning capabilities. Existing methods need to address key technical challenges such as real-time online measurement, internal defect detection, anti-interference processing, and adaptive optimization.

[0003] Traditional wind turbine blade measurement methods suffer from significant shortcomings in terms of accuracy and intelligence. Measurement equipment configurations are simple, and single-frequency radar or optical sensors cannot simultaneously acquire surface and internal information. Fixed signal filtering methods cannot adapt to varying noise characteristics in complex environments. Low phase calibration accuracy, coupled with equipment errors and environmental interference leading to signal distortion, results in simplistic feature recognition algorithms that fail to fully utilize the complementary information from multi-band signals. Linear parameter calculation models fail to accurately reflect the anisotropic characteristics of composite blades. Equipment adjustment mechanisms are lacking, with transmit power and receiver sensitivity settings relying on empirical values. Rigid continuous scanning strategies cannot optimize scanning paths and parameters based on blade condition. Limited pattern recognition capabilities make it difficult to extract early fault characteristics from dynamic data. Existing technologies necessitate the development of a fully intelligent measurement solution encompassing signal acquisition and condition assessment. Summary of the Invention

[0004] The purpose of this invention is to provide a millimeter-wave-based method for measuring wind turbine blades, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for measuring wind turbine blades based on millimeter waves, the method comprising:

[0006] Initialize the millimeter-wave radar system, set the multi-band transmission mode, transmit millimeter-wave signals to the wind turbine blades and receive the reflected signals, and convert the reflected signals into digital signal streams through an analog-to-digital converter;

[0007] Adaptive filtering and phase calibration are performed on the digital signal stream to eliminate environmental interference and equipment errors, generating enhanced signal data;

[0008] The enhanced signal data is analyzed to extract the surface contour features and internal structural features of the blade, and a blade feature dataset is constructed.

[0009] Based on the blade feature dataset, the size parameters and material property parameters of the blade are calculated to form a set of blade parameters;

[0010] Based on the set of blade parameters, the transmit power and receive sensitivity of the millimeter-wave radar are dynamically adjusted to achieve adaptive optimization of the measurement process;

[0011] The millimeter-wave radar is controlled to continuously scan the wind turbine blades according to the optimized parameters, and dynamic measurement data of time series is collected.

[0012] The system processes dynamic measurement data, uses pattern recognition algorithms to detect abnormal vibrations and structural changes in the blades, and generates blade condition measurement results.

[0013] Preferably, the step of converting the reflected signal into a digital signal stream via an analog-to-digital converter includes:

[0014] It receives the analog reflected signal from the millimeter-wave radar, uses a high-speed analog-to-digital converter to sample the signal at a preset sampling rate, and generates the original digital signal sequence.

[0015] The original digital signal sequence is quantized and encoded to convert it into a digital signal stream;

[0016] Verify the integrity and consistency of the digital signal stream to ensure that there is no signal loss or distortion.

[0017] Preferably, the adaptive filtering and phase calibration of the digital signal stream includes:

[0018] Collect noise samples from digital signal streams, estimate noise statistical characteristics, design adaptive filter coefficients, and apply filters to suppress high-frequency noise and low-frequency drift.

[0019] Phase deviation detection is performed on the filtered signal, the phase correction value is calculated, and the phase error caused by the signal propagation path is compensated.

[0020] The amplitude of the calibrated signal is normalized to generate enhanced signal data.

[0021] Preferably, the parsed and enhanced signal data includes:

[0022] Perform a short-time Fourier transform on the enhanced signal data to generate a time-spectrum graph;

[0023] Extract signal energy peaks and frequency components from the time-spectrum graph to identify reflection feature points of the blades;

[0024] Calculate the spatial relationships and temporal correlations between reflection feature points to construct a leaf feature dataset.

[0025] Preferably, the calculation of the blade's dimensional parameters and material property parameters includes:

[0026] A physical model is established based on the electromagnetic scattering theory of millimeter waves on the blade surface. As a scattering model, the initialization parameters of the scattering model include surface roughness and dielectric constant.

[0027] The leaf feature dataset is normalized and input into the scattering model. The gradient descent algorithm is used to iteratively optimize the model parameters and minimize the error between the model prediction and the actual feature data until the fitting residual is lower than the preset threshold, thus obtaining the calibrated scattering model.

[0028] Based on the calibrated scattering model, a geometric reconstruction algorithm is applied to process the feature point data in the reflected signal. Through point cloud registration and surface reconstruction techniques, the size parameters of the blade are calculated, including length, width, thickness distribution and curvature profile.

[0029] Simultaneously, the signal attenuation mode and phase shift are analyzed to derive the material's property parameters, including density and dielectric constant.

[0030] Integrate dimensional parameters and material property parameters to form a set of blade parameters.

[0031] Preferably, the dynamic adjustment of the millimeter-wave radar's transmit power and receive sensitivity includes:

[0032] Read the blade size and material properties from the blade parameter set, and calculate the required signal resolution and penetration depth;

[0033] Adjust the transmission frequency and power level of the millimeter-wave radar based on the calculation results;

[0034] Configure the receiver's gain settings and bandwidth parameters, test the adjusted parameter combinations, and ensure measurement coverage.

[0035] Preferably, the control of the millimeter-wave radar to continuously scan the wind turbine blades according to optimized parameters includes:

[0036] Set the scanning trajectory and scanning speed of the millimeter-wave radar, and start the continuous scanning mode according to the optimized parameters;

[0037] Real-time recording of intensity changes and timing information of reflected signals;

[0038] Simultaneously collect spatial position data and vibration data of the blades to generate dynamic measurement data in time series.

[0039] Preferably, the processing of dynamic measurement data includes:

[0040] Sliding window analysis is performed on dynamic measurement data to calculate signal statistical characteristics;

[0041] Compare current data with historical baseline data to identify abnormal signal patterns;

[0042] Locate the blade region corresponding to the abnormal mode and calculate the deformation and damage level.

[0043] Preferably, the step of extracting signal energy peaks and frequency components from the time-spectrum diagram and identifying reflection feature points of the blade includes:

[0044] The time-spectrum is binarized, and an adaptive threshold segmentation algorithm is used to distinguish between signal regions and background noise regions.

[0045] Clustering algorithms are applied within the signal region to identify connected components with high energy accumulation, and the center of each connected component is determined as a candidate reflection feature point.

[0046] Calculate the signal-to-noise ratio and frequency stability of each candidate reflection feature point, filter out candidate points with a signal-to-noise ratio lower than a preset threshold or that are unstable, and retain the effective blade reflection feature points;

[0047] Based on the effective timestamp and frequency information of the reflection feature points, their radial distance and velocity relative to the millimeter-wave radar are estimated.

[0048] Preferably, the identified signal anomaly pattern includes:

[0049] Extract time-domain and frequency-domain feature vectors from the sliding window analysis results, including signal energy distribution, spectral peaks, and transition characteristics;

[0050] The isolated forest algorithm is used to score anomalies in the feature vectors and calculate the anomaly probability of each data point.

[0051] Anomalies are classified into different levels based on their probability of occurrence, and the anomaly patterns are correlated with specific structural defect types by combining the results of blade area localization.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] The millimeter-wave radar system is initialized with a multi-band transmission mode, transmitting millimeter-wave signals to the wind turbine blades and receiving reflected signals. The reflected signals are then converted into a digital signal stream using an analog-to-digital converter (ADC). The multi-band transmission mode employs different frequency combinations, including Ka-band, Ku-band, and W-band, to balance penetration depth and resolution requirements. Millimeter-wave signals possess excellent atmospheric penetration and anti-interference characteristics, adapting to the wind turbine operating environment. The reflected signals contain comprehensive information from blade surface reflection and internal scattering, reflecting the blade's overall condition. The ADC utilizes a high sampling rate and high resolution configuration to ensure no signal detail is lost. The digital signal stream is stored in I / Q data format, preserving amplitude and phase information. Adaptive filtering and phase calibration are applied to the digital signal stream to eliminate environmental interference and equipment errors, generating enhanced signal data. Adaptive filtering employs LMS or RLS algorithms, dynamically adjusting filter coefficients based on noise characteristics. Phase calibration compensates for equipment transmission delay and antenna array errors, improving signal coherence. Environmental interference, including factors such as rain, snow, dust, and electromagnetic interference, is suppressed using multi-sensor fusion technology. Equipment error calibration employs reference targets and standard reflectors for system calibration. Enhancement processing improves the signal-to-noise ratio and spatial resolution, laying the foundation for feature extraction. Signal data is represented in a joint time-frequency distribution, preserving both time-domain and frequency-domain features.

[0054] The enhanced signal data is analyzed to extract surface contour features and internal structural features of the blade, constructing a blade feature dataset. Surface contour features reconstruct the blade's three-dimensional geometry using signal amplitude and phase information. Internal structural features utilize the penetration characteristics of multi-band signals to detect defects such as delamination, cracks, and water accumulation. Feature extraction employs microwave tomography and synthetic aperture radar techniques to improve spatial resolution. The blade feature dataset contains multi-dimensional information including geometric features, material features, and defect features. The dataset is organized in a hierarchical structure, supporting multi-granularity feature analysis. Based on the blade feature dataset, the blade's dimensional parameters and material property parameters are calculated, forming a blade parameter set. Dimensional parameters include geometric quantities such as blade length, chord length, thickness, and twist angle. Material property parameters reflect the composite material state through dielectric constant and loss tangent. Parameter calculation employs electromagnetic inverse scattering algorithms and machine learning models to improve inversion accuracy. The parameter set forms the basis for a digital twin model of the blade, supporting condition assessment and life prediction. The millimeter-wave radar's transmit power and receive sensitivity are dynamically adjusted based on the blade parameter set to achieve adaptive optimization of the measurement process. Transmit power adjustment optimizes signal penetration capability based on blade material and distance. The receiver sensitivity adjustment optimizes detection performance based on signal strength and ambient noise levels. Adaptive optimization employs a feedback control mechanism to adjust the system's operating state in real time. The optimization process balances measurement accuracy and energy efficiency, improving system economy. Through the synergistic effect of multi-band detection, adaptive processing, feature fusion, and parameter optimization, comprehensive monitoring of wind turbine blade status is achieved. Multi-band signals provide complementary information, adaptive processing improves signal quality, feature fusion enhances recognition capabilities, and parameter optimization improves measurement efficiency. This integrated approach significantly improves the accuracy and reliability of wind turbine blade measurements. Attached Figure Description

[0055] Figure 1 This is a schematic diagram illustrating the working principle of the millimeter-wave-based wind turbine blade measurement method described in this invention.

[0056] Figure 2 A flowchart illustrating the process of converting a reflected signal into a digital signal stream using an analog-to-digital converter;

[0057] Figure 3 A flowchart for analyzing the enhanced signal data;

[0058] Figure 4 A diagram showing the spatial distribution and energy intensity analysis of millimeter-wave reflection characteristic points on wind turbine blades;

[0059] Figure 5 This is a graph showing the dielectric constant and density characteristics of composite materials for wind turbine blades. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Please see Figure 1 This invention provides a method for measuring wind turbine blades based on millimeter waves. The method includes: initializing a millimeter-wave radar system, setting a multi-band transmission mode, transmitting millimeter-wave signals to the wind turbine blades and receiving reflected signals, converting the reflected signals into a digital signal stream using an analog-to-digital converter; performing adaptive filtering and phase calibration on the digital signal stream to eliminate environmental interference and equipment errors, generating enhanced signal data; parsing the enhanced signal data, extracting the surface contour features and internal structural features of the blades, and constructing a blade feature dataset; calculating the blade's size parameters and material property parameters based on the blade feature dataset to form a blade parameter set; dynamically adjusting the transmission power and receiving sensitivity of the millimeter-wave radar according to the blade parameter set to achieve adaptive optimization of the measurement process; controlling the millimeter-wave radar to continuously scan the wind turbine blades according to the optimized parameters, collecting dynamic measurement data in a time series; processing the dynamic measurement data, using a pattern recognition algorithm to detect abnormal vibrations and structural changes in the blades, and generating blade state measurement results.

[0062] Example 1: See Figure 2 In practical implementation, the millimeter-wave radar system receives analog reflected signals from wind turbine blades. These signals are sampled by a high-speed analog-to-digital converter at a preset sampling rate to generate an original digital signal sequence. In practice, the preset sampling rate is set according to the millimeter-wave signal frequency and the required resolution; for example, the sampling rate is no less than twice the highest frequency of the signal to avoid aliasing. The original digital signal sequence is then quantized and encoded into a digital signal stream. The quantization process uses a uniform quantization method to map the analog signal amplitude into discrete digital values, generating a digital signal stream in binary sequence form. The integrity and consistency of the digital signal stream are verified to ensure no signal loss or distortion. Verification methods include cyclic redundancy check (CRC) or parity check to detect errors during transmission. In some embodiments, integrity checks are performed by comparing the data packet length with the expected value, while consistency verification is completed by analyzing the signal timing continuity.

[0063] In practical implementation, noise samples are collected from the digital signal stream, noise statistical characteristics are estimated, adaptive filter coefficients are designed, and the filter is applied to suppress high-frequency noise and low-frequency drift. Noise samples are extracted from silent signal segments or background reference segments, and their statistical characteristics, including mean and variance, are used to initialize filter parameters. The adaptive filter uses the least mean square algorithm to update coefficients, and its update formula is as follows:

[0064] ,

[0065] in: Indicates the adaptive filter at time... coefficient vector, Indicates the adaptive step size parameter. Indicates time The error signal, Indicates time The input signal vector is used. In specific implementations, after the filter is applied, high-frequency noise and low-frequency drift components in the output signal are significantly suppressed. Phase deviation detection is performed on the filtered signal, and a phase correction value is calculated to compensate for phase errors caused by the signal propagation path. Phase deviation detection is achieved by comparing the phase difference between the reference signal and the received signal, and the phase correction value is calculated based on propagation delay and frequency offset. Amplitude normalization is performed on the calibrated signal to generate enhanced signal data. Amplitude normalization adjusts the signal amplitude to a uniform range, for example, by dividing by the maximum amplitude value. In some embodiments, the normalization process also considers the signal dynamic range to maintain feature integrity.

[0066] Optionally, multiple sampling windows can be set during noise sample acquisition to improve the accuracy of statistical estimation. Optionally, phase deviation detection can employ a multi-channel synchronization method to handle complex path effects. In phase deviation detection, the specific implementation of the multi-channel synchronization method includes configuring multiple independent receiving channels of the millimeter-wave radar system. These channels synchronously acquire signals based on the same clock source to simultaneously capture reflected signals from different paths from the wind turbine blades. Each receiving channel samples and buffers the filtered signal separately. Phase information from each channel is compared using cross-correlation analysis or phase interferometry to identify phase deviations caused by multipath propagation or environmental disturbances. The synchronous acquisition process ensures consistent signal timestamps, reducing errors caused by path delay differences. Subsequently, an average phase correction value is calculated based on the phase comparison results and applied to the signal data stream to compensate for phase distortion caused by complex path effects. This method utilizes inter-channel redundancy information to enhance detection robustness and adapt to dynamic environmental conditions in blade measurements. It is understood that the design of adaptive filter coefficients depends on real-time noise characteristics to adapt to environmental changes. It is understood that amplitude normalization contributes to the stability of subsequent feature extraction and avoids the impact of data scale differences on analysis results.

[0067] Example 2: See Figure 3 In the specific implementation, a short-time Fourier transform (SFT) is performed on the enhanced signal data to generate a time-spectrum graph. The SFT uses a sliding window method to process the signal, and a Hanning window is used as the window function to reduce spectral leakage. The transform result generates a two-dimensional time-spectrum graph matrix containing time, frequency, and energy information. Signal energy peaks and frequency components are extracted from the time-spectrum graph to identify blade reflection feature points. Energy peak detection is achieved through local maximum search, and frequency component analysis is based on the center frequency and bandwidth corresponding to the spectral peaks. Reflection feature points correspond to energy concentration areas in the time-spectrum graph. The spatial relationships and temporal correlations between reflection feature points are calculated to construct a blade feature dataset. Spatial relationships are described by Euclidean distance and angle between feature points, and temporal correlations are calculated using time series analysis to determine the periodicity or trend of feature point occurrences. The dataset includes the coordinates, energy values, and frequency attributes of the feature points.

[0068] In practical implementation, the time-spectrum image is binarized, and an adaptive threshold segmentation algorithm is used to distinguish signal regions from background noise regions. The adaptive threshold is calculated based on local image patches, such as using the mean method, to convert the time-spectrum image into a binary image, where high-value regions represent signals and low-value regions represent noise. Within the signal regions, a clustering algorithm is applied to identify connected components with high energy accumulation. The center of each connected component is determined as a candidate reflection feature point. The clustering algorithm uses the density-based DBSCAN method, where connected components consist of adjacent high-energy pixels, and the center coordinates are obtained by calculating the weighted average of the pixels within the connected component. The signal-to-noise ratio (SNR) and frequency stability of each candidate reflection feature point are calculated. Candidate points with an SNR below a preset threshold or unstable are filtered out, retaining valid blade reflection feature points. The SNR is calculated using the formula:

[0069] ,

[0070] in: Indicates the signal-to-noise ratio. Indicates the signal energy of candidate points. This represents the average energy of the background noise region. The standard deviation of energy in the background noise region is represented. Frequency stability is evaluated by analyzing the variance of frequency changes at candidate points in different time windows. The preset threshold is set according to the system noise level. Based on the timestamps and frequency information of effective reflection feature points, their radial distance and velocity relative to the millimeter-wave radar are estimated. The radial distance is calculated using the signal propagation time delay, and the velocity is derived based on the Doppler frequency shift. In some embodiments, the timestamp is synchronized with the radar system clock, and the frequency information is extracted from the short-time Fourier transform results.

[0071] In some embodiments, the adaptive thresholding segmentation algorithm can combine global and local thresholds to handle non-uniform backgrounds. In a specific implementation of the adaptive thresholding segmentation algorithm, the time-spectrum image is divided into multiple local image blocks, and a threshold is calculated for each local image block. Simultaneously, a global threshold for the entire time-spectrum image is calculated as a benchmark. The local and global thresholds are combined using a weighted averaging or dynamic adjustment strategy to adapt to uneven background noise distribution and ensure accurate differentiation between signal regions and background noise regions. In some embodiments, the parameter settings of the clustering algorithm are dynamically adjusted based on the resolution of the time-spectrum image. Optionally, a weighting factor can be introduced in the signal-to-noise ratio calculation to emphasize the importance of specific frequency bands. Optionally, radial distance estimation can compensate for atmospheric attenuation and device delay. It is understood that adaptive thresholding segmentation helps adapt to noise variations in different environments. It is understood that the clustering algorithm can effectively identify multiple reflection sources in complex scenes.

[0072] See Figure 4 This figure uses X and Y coordinates to display the spatial location of reflection feature points within the blade region, with the color gradient on the right representing the energy of these feature points. It is a direct visualization of the results of extracting blade reflection feature points through a short-time Fourier transform of the enhanced signal. The distribution in the figure shows that different gray levels of the feature points correspond to energy differences: bright points have high energy, representing strong reflection areas on the blade surface or internal structures; dark points have low energy, corresponding to weak reflection areas. The spatial relationships and temporal correlations of these feature points are core to constructing the blade feature dataset, providing crucial information for subsequent blade contour reconstruction and internal defect localization. This figure fully presents the spatial distribution and energy attributes of feature points after millimeter-wave signal analysis, serving as a key visualization tool for identifying effective reflection feature points from the time-spectrum diagram and supporting multi-dimensional feature analysis of the blade.

[0073] Example 3: In specific implementation, a physical model is established based on the electromagnetic scattering theory of millimeter waves on the blade surface. This model serves as the initialization parameter for the scattering model, including surface roughness and dielectric constant. The surface roughness parameter describes the microscopic geometry of the blade surface, while the dielectric constant parameter reflects the electromagnetic properties of the blade material. The scattering model is constructed based on Kirchhoff's approximation or physical optics theory to simulate the interaction between millimeter waves and the blade surface. The blade feature dataset is normalized and input into the scattering model. The gradient descent algorithm iteratively optimizes the model parameters, minimizing the error between the model's predicted values ​​and the actual feature data, until the fitting residual is below a preset threshold, thus obtaining the calibrated scattering model. Feature normalization scales the feature values ​​in the dataset to a uniform range, for example, by using a linear transformation to distribute the values ​​between zero and one. The gradient descent algorithm updates the model parameters by calculating the gradient of the loss function. The fitting residual is used to evaluate the model's prediction accuracy, and its calculation formula is as follows:

[0074] ,

[0075] in: Represents the fitting residual. Indicates the number of feature data points. Indicates the first One actual feature data, Indicates the scattering model for the first The predicted values ​​of each data point are set with a preset threshold based on the measurement accuracy requirements. Iteration terminates when the fitting residual falls below this threshold. Based on the calibrated scattering model, a geometric reconstruction algorithm is applied to process the feature point data in the reflected signal. Through point cloud registration and surface reconstruction techniques, the blade's dimensional parameters, including length, width, thickness distribution, and curvature profile, are calculated. The geometric reconstruction algorithm is based on triangulation or implicit surface methods. Point cloud registration aligns the feature point cloud using an iterative nearest-point algorithm. Surface reconstruction uses Poisson reconstruction or moving least squares to generate a smooth surface, and the dimensional parameters are extracted from the reconstructed surface. Simultaneously, the signal attenuation mode and phase shift are analyzed to derive material property parameters, including density and dielectric constant. The signal attenuation mode is described by the amplitude attenuation coefficient, the phase shift is calculated by the propagation path difference, and the material property parameters are derived using the interaction between electromagnetic waves and matter. The dimensional parameters and material property parameters are integrated to form a blade parameter set, stored in a structured data format containing numerical and unit information.

[0076] In some embodiments, the initialization parameters of the scattering model can be loaded from a predefined material library to accelerate the model convergence process. In some embodiments, the geometric reconstruction algorithm can be combined with multi-view data fusion to improve the accuracy of size calculation. Optionally, feature normalization processing can employ quantile normalization methods to handle non-uniformly distributed data. Optionally, point cloud registration can introduce feature descriptor matching to improve registration robustness.

[0077] See Figure 5This figure uses material samples as the horizontal axis, simultaneously presenting the quantitative characteristics of two key material properties: the dielectric constant on the left vertical axis reflects the material's response to millimeter-wave electromagnetic waves, while the density on the right vertical axis reflects the material's mass distribution characteristics. These two are core data carriers for deriving blade material property parameters based on millimeter-wave electromagnetic scattering theory. As can be seen from the distribution in the figure, sample 3 has both its dielectric constant and density at their peaks, indicating that this sample performs exceptionally well in electromagnetic interaction and physical mass density, providing a typical reference for modeling the penetration and reflection of millimeter-wave signals. Sample 1 has the lowest dielectric constant and density, representing the other extreme of material properties; the parameters of samples 2, 4, and 5 exhibit a gradient distribution. This difference provides a multi-dimensional basis for constructing a blade feature dataset and calibrating the electromagnetic scattering model. The dielectric constant directly affects the propagation and scattering behavior of millimeter waves inside the blade, while density is related to the stability and damage susceptibility of the blade structure. The synergistic analysis of these two is a key step in realizing the inversion of blade material and structural defects from millimeter-wave signals, contributing to the improved accuracy of subsequent blade size parameter calculations and anomaly pattern recognition.

[0078] Example 4: In a specific implementation, blade size and material property parameters are read from the blade parameter set. The blade size parameters include length, width, and thickness distribution, while the material property parameters include density and dielectric constant. These parameters are loaded from the storage medium in a structured data format. The required signal resolution and penetration depth are calculated. The required signal resolution is derived based on the minimum characteristic size of the blade and the material dielectric constant, using the formula:

[0079] ,

[0080] in: Indicates the required signal resolution. This represents the resolution scaling factor. Indicates the minimum characteristic size of the blade. The relative permittivity of the material is represented by the penetration depth, calculated based on the maximum blade thickness and material attenuation characteristics to ensure electromagnetic waves can penetrate the internal structure of the blade. The transmission frequency and power level of the millimeter-wave radar are adjusted according to the calculation results. A higher frequency band is selected to improve spatial resolution, or a lower frequency band to enhance penetration capability. The power level is dynamically set based on the estimated signal attenuation. The receiver's gain and bandwidth parameters are configured. The gain setting is adjusted based on the expected signal strength, and the bandwidth parameter is matched to the transmission spectrum to optimize the signal-to-noise ratio. Referring to Table 1, the adjusted parameter combinations are tested to ensure measurement coverage. The testing process includes transmitting a calibration signal and analyzing the echo intensity distribution.

[0081] Table 1: Millimeter-wave radar parameter adjustment mapping table

[0082]

[0083] In practice, the resolution scaling factor κ is set according to the system accuracy requirements, and the penetration depth is calculated using an electromagnetic wave propagation model that combines material density and dielectric constant. The transmit frequency is adjusted with reference to a predefined frequency-resolution curve, and the power level is set considering path loss and material absorption. The receiver gain is set using an automatic gain control algorithm, and the bandwidth parameter is configured according to the signal modulation type. When testing the adjusted parameter combination, it is verified whether the measurement coverage area covers the entire blade area, and the absence of blind spots is confirmed by a signal strength map.

[0084] In some embodiments, the resolution scaling factor κ can be selected from 0.1 to 0.5 based on the type of object being detected. In some embodiments, a safety factor is incorporated into the penetration depth calculation to account for material inhomogeneity. Optionally, the transmit frequency adjustment can be optimized in conjunction with an environmental interference assessment. Optionally, the receiver bandwidth parameter can be adaptively varied with the signal dynamic range. It is understood that the required signal resolution calculation depends on the accuracy of the blade geometry.

[0085] Example 5: In specific implementation, the scanning trajectory and scanning speed of the millimeter-wave radar are set, and a continuous scanning mode is started according to the optimized parameters. The scanning trajectory is preset as a helix or grid path based on the blade geometry. The scanning speed is dynamically adjusted according to the blade size and measurement accuracy requirements. The continuous scanning mode maintains the continuous operation of the radar transmitting and receiving circuits. The intensity changes and timing information of the reflected signal are recorded in real time. The intensity changes are stored in decibels, and the timing information includes the signal arrival timestamp and duration sequence. Spatial position data and vibration data of the blade are collected synchronously to generate dynamic measurement data in time series. Spatial position data is obtained through a global positioning system or inertial measurement unit, and vibration data is analyzed through accelerometers or radar Doppler signals. The dynamic measurement data is organized into a multi-dimensional array in chronological order.

[0086] In practice, a sliding window analysis is performed on the dynamic measurement data to calculate the signal statistical characteristics. The size of the sliding window is set according to the signal sampling rate and characteristic period. The statistical characteristics include the mean, variance, and peak value. The root mean square value of the signal within the window is calculated using the following formula:

[0087] ,

[0088] in: This represents the root mean square value of the signal within the sliding window. Indicates the number of signal samples within the window. Indicates the first The amplitude of each signal sample is measured. Current data is compared with historical baseline data to identify signal anomaly patterns. Historical baseline data comes from measurement records under fault-free blade conditions. Anomaly patterns are detected by deviation thresholds; if statistical characteristics exceed a preset range, the anomaly pattern is identified. The blade region corresponding to the anomaly pattern is located, and the deformation and damage degree are calculated. Location is based on the signal source angle and distance information. Deformation is obtained by comparing the actual profile with the baseline model, and damage degree is assessed by the defect area ratio or depth.

[0089] In practical implementation, time-domain and frequency-domain feature vectors are extracted from the sliding window analysis results, including signal energy distribution, spectral peaks, and transition characteristics. The time-domain feature vector includes mean, variance, and zero-crossing rate, while the frequency-domain feature vector uses Fast Fourier Transform to calculate the spectral envelope and dominant frequency. The Isolation Forest algorithm is applied to score anomalies in the feature vectors, calculating the anomaly probability for each data point. The Isolation Forest algorithm constructs multiple isolated trees, using path length to measure the degree of anomaly at each data point. The anomaly probability is mapped to a zero-to-one interval using a scoring function. Anomaly levels are classified based on the anomaly probability, and combined with the blade region localization results, anomaly patterns are correlated with specific structural defect types. Anomaly levels are divided into low, medium, and high, corresponding to different response strategies. Structural defect types include cracks, delamination, and corrosion.

[0090] In some embodiments, the sliding window size can adaptively change with the blade rotation speed to maintain analytical consistency. In some embodiments, the anomaly level threshold can be dynamically adjusted according to the operating environment. Optionally, feature vector extraction may include higher-order statistics such as skewness and kurtosis. Optionally, the parameters of the isolated forest algorithm can be optimized through cross-validation. In the parameter optimization of the isolated forest algorithm, cross-validation optimization is implemented using historical benchmark data, which is divided into multiple mutually exclusive subsets. Each subset is used as a validation set, and the remaining subsets are used as training sets. Multiple isolated forest models are trained for different parameter combinations. Each model learns the distribution pattern of feature vectors on the training set and calculates the accuracy index of anomaly scores on the validation set. By comparing the performance of each model on the validation set, the parameter combination that makes the anomaly probability calculation most stable is selected. Finally, the optimized parameters are applied to the anomaly pattern recognition process of dynamic measurement data.

[0091] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0092] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A millimeter wave-based method of measuring a fan blade, the method comprising: The method implements the following steps: Initialize the millimeter wave radar system, set the multi-band transmission mode, transmit millimeter wave signals to the fan blades and receive reflected signals, and convert the reflected signals into a digital signal stream through an analog-to-digital converter; Adaptive filtering and phase calibration are performed on the digital signal stream to eliminate environmental interference and equipment errors, and enhanced signal data is generated; Parse the enhanced signal data, extract the surface profile features and internal structure features of the blades, and construct a blade feature dataset; Based on the blade feature dataset, calculate the size parameters and material attribute parameters of the blades to form a blade parameter set; According to the blade parameter set, dynamically adjust the transmission power and reception sensitivity of the millimeter wave radar to achieve adaptive optimization of the measurement process; Control the millimeter wave radar to continuously scan the fan blades according to the optimized parameters to collect time series of dynamic measurement data; Process the dynamic measurement data and use pattern recognition algorithms to detect abnormal vibration and structural changes of the blades to generate blade state measurement results; The calculation of the size parameters and material attribute parameters of the blades includes: Based on the electromagnetic scattering theory of millimeter waves on the surface of the blade, a physical model is established as a scattering model, and the initialization of the scattering model parameters includes surface roughness and dielectric constant; Perform feature normalization on the blade feature dataset, input the scattering model, and use the gradient descent algorithm to iteratively optimize the model parameters to minimize the error between the model prediction and the actual feature data until the fitting residual is below the preset threshold to obtain the calibrated scattering model; Based on the calibrated scattering model, apply the geometric reconstruction algorithm to process the feature point data in the reflected signal, calculate the size parameters of the blade including length, width, thickness distribution and curvature profile through point cloud registration and surface reconstruction technology; At the same time, analyze the signal attenuation pattern and phase shift to derive the material attribute parameters including density and dielectric constant; Integrate the size parameters and material attribute parameters to form the blade parameter set.

2. The millimeter-wave-based fan blade measurement method of claim 1, wherein, The conversion of the reflected signal into a digital signal stream through an analog-to-digital converter includes: Receive the analog reflected signal of the millimeter wave radar, use a high-speed analog-to-digital converter to sample the signal at a preset sampling rate to generate an original digital signal sequence; Quantize and encode the original digital signal sequence to convert it into a digital signal stream; Verify the integrity and consistency of the digital signal stream to ensure that the signal is not lost or distorted.

3. The millimeter-wave-based fan blade measurement method of claim 1, wherein, The adaptive filtering and phase calibration of the digital signal stream includes: Collect noise samples in the digital signal stream, estimate the noise statistical characteristics, design adaptive filter coefficients, and apply the filter to suppress high-frequency noise and low-frequency drift; Detect the phase deviation of the filtered signal, calculate the phase correction value, and compensate for the phase error caused by the signal propagation path; Amplitude normalization is performed on the calibrated signal to generate enhanced signal data.

4. The millimeter-wave-based fan blade measurement method of claim 1, wherein, The analysis of the enhanced signal data includes: Perform short-time Fourier transform on the enhanced signal data to generate a time-frequency spectrum; Extract signal energy peaks and frequency components from the time-frequency spectrum to identify blade reflection feature points; Calculate the spatial relationship and time sequence correlation between the reflection feature points to construct a blade feature dataset.

5. The millimeter-wave-based fan blade measurement method of claim 1, wherein, The dynamic adjustment of the transmission power and receiving sensitivity of the millimeter wave radar includes: Reading the blade size and material properties from the blade parameter set, calculating the required signal resolution and penetration depth; Adjusting the transmission frequency and power level of the millimeter wave radar according to the calculation results; Configuring the gain setting and bandwidth parameters of the receiver, testing the adjusted parameter combination to ensure the measurement coverage.

6. The millimeter-wave-based fan blade measurement method of claim 1, wherein, The control of the millimeter wave radar to continuously scan the fan blades according to the optimized parameters includes: Setting the scanning trajectory and scanning speed of the millimeter wave radar, and starting the continuous scanning mode according to the optimized parameters; Real-time recording of the intensity change and timing information of the reflected signal; Synchronously collecting the spatial position data and vibration data of the blades to generate time-series dynamic measurement data.

7. The millimeter-wave-based fan blade measurement method of claim 1, wherein, The processing of dynamic measurement data includes: Sliding window analysis of dynamic measurement data to calculate signal statistical features; Comparing the current data with historical baseline data to identify signal anomaly patterns; Locating the blade area corresponding to the anomaly pattern, calculating the deformation and damage degree.

8. The millimeter-wave-based fan blade measurement method of claim 4, wherein, The extraction of signal energy peaks and frequency components from the time-frequency spectrum to identify blade reflection feature points includes: Binaryzation of the time-frequency spectrum, using adaptive threshold segmentation algorithm to distinguish signal area and background noise area; Applying clustering algorithm in the signal area to identify connected domains with high energy accumulation, and determining the center of each connected domain as a candidate reflection feature point; Calculating the signal-to-noise ratio and frequency stability of each candidate reflection feature point, filtering out candidate points with signal-to-noise ratio below the preset threshold or unstable, and retaining effective blade reflection feature points; Based on the timestamp and frequency information of the effective reflection feature points, estimating their radial distance and speed relative to the millimeter wave radar.

9. The millimeter-wave-based fan blade measurement method of claim 7, wherein, The identification of signal anomaly patterns includes: Extracting time-domain and frequency-domain feature vectors from the sliding window analysis results, including signal energy distribution, spectral peak and transition characteristics; Applying Isolation Forest algorithm to anomaly score the feature vectors, calculating the anomaly probability of each data point; According to the anomaly probability, dividing the anomaly level, and combining with the blade area positioning result, associating the anomaly pattern with the specific structure defect type.

Images