Fan blade deformation monitoring method and system based on three-dimensional fast point cloud model

Abstract

The present invention relates to the technical field of fan blade deformation monitoring, and proposes a fan blade deformation monitoring method and system based on a three-dimensional fast point cloud model, comprising the following steps: obtaining monitoring data of fan blades through multiple types of sensors; constructing an environmental factor compensation model based on a deep learning network; performing wind force compensation and temperature compensation on the three-dimensional point cloud data through the environmental factor compensation model to obtain compensated point cloud data; optimizing the compensated point cloud data to obtain optimized point cloud data; constructing a fan three-dimensional blade model based on the optimized point cloud data, and obtaining fan blade deformation data by comparing the fan three-dimensional blade model with the blade design model; and analyzing the fan blade deformation data in real time, and outputting a deformation warning signal when the deformation of the fan blade exceeds a preset threshold. The present invention realizes accurate monitoring and real-time warning of fan blade deformation, improves the accuracy and real-time performance of monitoring, and ensures the safe and stable operation of the fan.

Description

Technical Field

[0001] The present invention relates to the technical field of fan blade deformation monitoring, and in particular to a fan blade deformation monitoring method and system based on a three-dimensional fast point cloud model. Background Art

[0002] Wind power generation is an important part of the new energy field, and its power generation efficiency and safety directly affect the stability of energy supply. Wind turbine blades operate for a long time in a complex and changeable natural environment. They must not only withstand the dynamic changes of wind loads, but also adapt to the combined influence of environmental factors such as temperature and humidity, which makes the blades prone to deformation, fatigue damage and other problems. If these problems cannot be discovered and handled in time, it will seriously affect the power generation efficiency of the wind turbine and even lead to catastrophic failures.

[0003] The data collection method of the existing technology adopts fixed frequency sampling, which cannot adaptively adjust the sampling strategy according to the dynamic changes of the blade state, resulting in missing important data at critical moments or generating a large amount of redundant data during the stable period, reducing the accuracy and real-time performance of wind turbine blade deformation monitoring. Summary of the invention

[0004] In view of this, the present invention proposes a fan blade deformation monitoring method and system based on a three-dimensional fast point cloud model, which solves the problem that the prior art uses fixed frequency sampling and cannot adaptively adjust the sampling strategy according to the dynamic changes of the blade state, resulting in missing important data at critical moments or generating a large amount of redundant data during the stable period, thereby reducing the accuracy and real-time performance of fan blade deformation monitoring.

[0005] The technical solution of the present invention is implemented as follows: On the one hand, the present invention provides a method for monitoring the deformation of a fan blade based on a three-dimensional fast point cloud model, comprising the following steps:

[0006] Acquire monitoring data of wind turbine blades through multiple types of sensors, wherein the monitoring data includes three-dimensional point cloud data, temperature data and wind speed data;

[0007] Constructing an environmental factor compensation model based on a deep learning network;

[0008] By using the environmental factor compensation model, according to the temperature data and the wind speed data, the three-dimensional point cloud data is subjected to wind force compensation and temperature compensation to obtain compensated point cloud data;

[0009] Performing optimization processing on the compensated point cloud data, wherein the optimization processing includes noise filtering, data alignment and dynamic updating to obtain optimized point cloud data;

[0010] Building a three-dimensional fan blade model based on the optimized point cloud data, and obtaining fan blade deformation data by comparing the three-dimensional fan blade model with a blade design model;

[0011] The fan blade deformation data is analyzed in real time, and when the deformation amount of the fan blade exceeds a preset threshold, a deformation warning signal is output.

[0012] On the basis of the above technical solution, preferably, the monitoring data of the wind turbine blades is obtained by using multiple types of sensors, and the monitoring data includes three-dimensional point cloud data, temperature data and wind speed data, specifically including:

[0013] Multiple monitoring stations are set up around the wind turbine blades. LiDAR, temperature sensor and wind speed sensor are set up at each monitoring station to form a multi-station collaborative monitoring network through all monitoring stations.

[0014] The monitoring data of the wind turbine blades is collected through the multi-site collaborative monitoring network, and the monitoring data is transmitted using a wireless sensor network. The wireless sensor network includes multiple wireless sensor nodes and a data aggregation node. Each monitoring site corresponds to a wireless sensor node. The wireless sensor node adopts an adaptive sampling frequency mechanism to dynamically adjust the sampling frequency of the monitoring data according to the change rate of the monitoring data.

[0015] On the basis of the above technical solution, preferably, the environmental factor compensation model is constructed based on the deep learning network, specifically including:

[0016] A dual-branch deep neural network model including a temperature compensation subnetwork and a wind compensation subnetwork is constructed, wherein the temperature compensation subnetwork includes a feature extraction layer, a physical constraint layer, and a compensation coefficient generation layer; the feature extraction layer adopts a multi-layer perceptron structure to extract the spatiotemporal characteristics of temperature data; the physical constraint layer is used to calculate the temperature strain coefficient; the compensation coefficient generation layer generates a temperature compensation coefficient matrix of three-dimensional point cloud data based on the temperature strain coefficient;

[0017] The calculation formula of the temperature strain coefficient is:

[0018] ε1=α1×ΔT1×(1+β1×sin(ω1t1))×exp(-γ1×d1);

[0019] Among them, ε1 is the temperature strain coefficient, α1 is the linear thermal expansion coefficient of the fan blade material, ΔT1 is the temperature change, β1 is the temperature nonlinear correction coefficient, ω1 is the temperature cycle change frequency, t1 is the first time parameter, exp(·) is the exponential function, γ1 is the temperature spatial attenuation coefficient, and d1 is the distance from the measurement point to the blade surface;

[0020] The calculation formula of the temperature compensation coefficient matrix is:

[0021]

[0022]

[0023] Where M is the temperature compensation coefficient matrix, W is the spatial weight matrix, ⊙ is the Hadamard product, ε1 is the temperature strain coefficient, I is the unit matrix, σ1 is the temperature gradient coefficient, ▽T2 is the temperature gradient matrix, δ1 is the conduction attenuation coefficient, ρ is the spatial attenuation rate, R is the distance matrix, exp(·) is the exponential function, P is the current point cloud coordinate matrix, P ref is the reference point cloud coordinate matrix, τ1 is the weight adjustment parameter, and ||·||2 is the Euclidean distance norm;

[0024] Pre-training the dual-branch deep neural network through simulation data to obtain an initial training environmental factor compensation model, acquiring historical monitoring data, fine-tuning the initial training environmental factor compensation model based on the historical monitoring data to obtain an environmental factor compensation model to be optimized, and continuously optimizing the model parameters of the environmental factor compensation model to be optimized by online learning to obtain an environmental factor compensation model; adopting an adaptive learning rate adjustment mechanism during the training process to dynamically adjust the learning rate of the model according to the performance of the validation set;

[0025] The calculation formula of the environmental factor compensation model is:

[0026]

[0027] Among them, P comp is the compensated point cloud data, P orig is the input original point cloud data, ⊙ is the Hadamard product, M is the temperature compensation coefficient matrix, N is the wind compensation coefficient matrix, is the time-varying adaptive weight function, and C is the spatial constraint matrix.

[0028] On the basis of the above technical solution, preferably, the environmental factor compensation model is used to perform wind compensation and temperature compensation on the three-dimensional point cloud data according to the temperature data and the wind speed data to obtain compensated point cloud data, which specifically includes:

[0029] A sliding time window mechanism is used to process temperature data and wind speed data, and the window length is adaptively adjusted according to the data change rate of the temperature data and wind speed data; feature extraction and normalization are performed on the temperature data and wind speed data in each time window to generate a feature vector; the feature vector is input into the dual-branch network of the environmental factor compensation model to obtain the output of the temperature compensation branch and the wind compensation branch respectively; the weight coefficients of the two branches are calculated through the attention mechanism, and they are fused based on the adaptive weight function to obtain a comprehensive compensation coefficient matrix;

[0030] An iterative optimization algorithm based on local consistency constraints is used for point cloud compensation. Each iteration includes: calculating the residual between the current point cloud data and the local reference model; updating the comprehensive compensation coefficient matrix based on the residual; correcting the current point cloud data using the updated comprehensive compensation coefficient matrix; calculating the local curvature change of the corrected point cloud data; and terminating the iterative process when the curvature change of two consecutive iterations is less than a preset threshold.

[0031] On the basis of the above technical solution, preferably, the compensation point cloud data is optimized, and the optimization process includes noise filtering, data alignment and dynamic update to obtain optimized point cloud data, specifically including:

[0032] Noise filtering is performed on the compensated point cloud data, including: calculating the local neighborhood features of each point, the local neighborhood features including point density, curvature and normal vector, setting a density threshold and a curvature threshold based on statistical feature analysis, removing low-density points and high-curvature abnormal points based on the density threshold and the curvature threshold, and performing probability filtering based on Gaussian distribution on the remaining point cloud data;

[0033] The point cloud data after noise filtering is aligned and dynamically updated, including: using an alignment method based on an iterative closest point algorithm to calculate the rigid body transformation matrix between the current point cloud data and the reference model, and based on real-time monitoring data, using a Kalman filter algorithm to dynamically update the point cloud data, correcting the time series error of the point cloud, and obtaining optimized point cloud data.

[0034] On the basis of the above technical solution, preferably, the three-dimensional blade model of the wind turbine is constructed based on the optimized point cloud data, and the deformation data of the wind turbine blade is obtained by comparing the three-dimensional blade model of the wind turbine with the blade design model, which specifically includes:

[0035] A point cloud fitting method based on B-spline surface is used to extract key feature points of optimized point cloud data, and surface reconstruction is performed based on the key feature points to construct a three-dimensional blade model of the wind turbine;

[0036] The calculation formula for the surface reconstruction of the three-dimensional fan blade model is:

[0037]

[0038] Among them, S(u,w) is the surface reconstructed in the u and w directions, N a,p(u) is the first B-spline basis function, N b,q(w) is the second B-spline basis function, P a,b is the control point, k4 is the number of control points in the u direction minus 1, k5 is the number of control points in the w direction minus 1, p is the order of the first B-spline, and q is the order of the second B-spline;

[0039] A comparative analysis method based on point cloud distance is used to align the wind turbine 3D blade model with the blade design model, calculate the deformation of each point, and generate a deformation thermal map of the blade based on the deformation distribution.

[0040] The calculation formula of the deformation is:

[0041] Δd c =||p c -q c ||+α2·|κ c -κ ref |;

[0042] Where, Δd c is the cth point p c The deformation, q c is the position of the corresponding c-th point in the design model, κ c and κ ref are the local curvature functions of the current model and the design model respectively, and α2 is the curvature weight coefficient;

[0043] The calculation formula of the deformation thermodynamic map is:

[0044]

[0045] Where H(e,f) is the deformation value of the point (e,f) in the thermal diagram, Δd g For point p g The deformation amount, σ3 is the spatial scale parameter, and k6 is the number of points in the deformation thermodynamic map.

[0046] On the basis of the above technical solution, preferably, the real-time analysis of the fan blade deformation data and outputting a deformation warning signal when the fan blade deformation exceeds a preset threshold value specifically includes:

[0047] Performing multi-dimensional analysis on the deformation data of the fan blades to obtain blade deformation characteristic parameters, wherein the multi-dimensional analysis includes deformation distribution analysis, deformation trend analysis and deformation rate analysis;

[0048] An early warning judgment is performed based on the blade deformation characteristic parameters. When any characteristic parameter of the blade deformation characteristic parameters exceeds a corresponding preset threshold, an early warning level is determined and a corresponding deformation early warning signal is output.

[0049] In a second aspect, the present invention further provides a fan blade deformation monitoring system based on a three-dimensional fast point cloud model, the system comprising:

[0050] A data acquisition module, used to collect monitoring data of wind turbine blades through multiple types of sensors, wherein the monitoring data includes three-dimensional point cloud data, temperature data and wind speed data;

[0051] Compensation model module, used to build an environmental factor compensation model based on a deep learning network;

[0052] A point cloud compensation module, used to perform wind force compensation and temperature compensation on the three-dimensional point cloud data according to the temperature data and the wind speed data through the environmental factor compensation model to obtain compensated point cloud data;

[0053] A point cloud optimization module, used for optimizing the compensated point cloud data, wherein the optimization includes noise filtering, data alignment and dynamic updating, to obtain optimized point cloud data;

[0054] A deformation comparison module, used for constructing a three-dimensional fan blade model based on the optimized point cloud data, and obtaining fan blade deformation data by comparing the three-dimensional fan blade model with a blade design model;

[0055] The deformation warning module is used to analyze the deformation data of the fan blades in real time and output a deformation warning signal when the deformation amount of the fan blades exceeds a preset threshold.

[0056] In a third aspect, the present invention further provides an electronic device, comprising: at least one processor, at least one memory, a communication interface and a bus;

[0057] Among them, the processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement steps such as a method for monitoring wind blade deformation based on a three-dimensional fast point cloud model.

[0058] In a fourth aspect, the present invention further provides a computer-readable storage medium storing computer instructions, wherein the computer instructions enable a computer to implement steps of a method for monitoring deformation of a wind turbine blade based on a three-dimensional fast point cloud model.

[0059] The fan blade deformation monitoring method and system based on a three-dimensional fast point cloud model of the present invention has the following beneficial effects compared with the prior art:

[0060] (1) Through the collaborative collection of three-dimensional point cloud data, temperature data, and wind speed data of wind turbine blades by multiple types of sensors, the environmental factor compensation model constructed by the deep learning network is used to compensate for wind force and temperature of the point cloud data. The quality and accuracy of the point cloud data are improved through optimization processing, which enables accurate monitoring and real-time warning of wind turbine blade deformation, improves the accuracy and real-time performance of monitoring, and ensures the safe and stable operation of the wind turbine.

[0061] (2) By arranging multiple monitoring stations in a ring around the wind turbine blades, combining the adaptive sampling mechanism of the wireless sensor network, and adopting a deep neural network model with two branches, temperature compensation and wind force compensation, and introducing physical constraints of the temperature strain coefficient and wind force compensation coefficient, all-round high-precision monitoring of the blades is achieved. On the basis of ensuring the comprehensiveness and reliability of data collection, compensation for the influence of environmental factors is further achieved, thereby improving the accuracy of point cloud data;

[0062] (3) By adopting a multi-level noise filtering mechanism based on local neighborhood features and an iterative nearest point algorithm for data optimization, the Kalman filter is combined to realize the dynamic update of point cloud data, and the B-spline surface point cloud fitting method is used to reconstruct the three-dimensional model of the blade. On the basis of ensuring the quality and time continuity of the point cloud data, a high-precision blade model is constructed. By calculating the point cloud distance and generating a deformation heat map, the deformation distribution of each area of ​​the blade is intuitively displayed, thereby improving the visualization effect of the deformation monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 A flow chart of a method for monitoring deformation of a fan blade based on a three-dimensional fast point cloud model according to the present invention;

[0065] Figure 2 This is a structural diagram of a fan blade deformation monitoring system based on a three-dimensional fast point cloud model according to the present invention. DETAILED DESCRIPTION

[0066] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] See also Figure 1 The present invention provides a method for monitoring fan blade deformation based on a three-dimensional fast point cloud model, comprising the following steps:

[0068] Acquire monitoring data of wind turbine blades through multiple types of sensors, wherein the monitoring data includes three-dimensional point cloud data, temperature data and wind speed data, wherein the multiple types of sensors include a laser radar, a temperature sensor and a wind speed sensor, wherein the laser radar is used to collect three-dimensional point cloud data, and the temperature sensor and the wind speed sensor are used to collect temperature data and wind speed data;

[0069] Constructing an environmental factor compensation model based on a deep learning network;

[0070] By using the environmental factor compensation model, according to the temperature data and the wind speed data, the three-dimensional point cloud data is subjected to wind force compensation and temperature compensation to obtain compensated point cloud data;

[0071] Performing optimization processing on the compensated point cloud data, wherein the optimization processing includes noise filtering, data alignment and dynamic updating to obtain optimized point cloud data;

[0072] Building a three-dimensional fan blade model based on the optimized point cloud data, and obtaining fan blade deformation data by comparing the three-dimensional fan blade model with a blade design model;

[0073] The fan blade deformation data is analyzed in real time, and when the deformation amount of the fan blade exceeds a preset threshold, a deformation warning signal is output.

[0074] Specifically, this embodiment uses multiple types of sensors to collaboratively collect three-dimensional point cloud data, temperature data, and wind speed data of wind turbine blades, combines the environmental factor compensation model constructed by the deep learning network, performs wind force and temperature compensation on the point cloud data, and improves the quality and accuracy of the point cloud data through optimization processing, thereby achieving accurate monitoring and real-time early warning of wind turbine blade deformation, improving the accuracy and real-time performance of monitoring, and ensuring the safe and stable operation of the wind turbine.

[0075] The monitoring data of the wind turbine blades is obtained by using multiple types of sensors, and the monitoring data includes three-dimensional point cloud data, temperature data and wind speed data, specifically including:

[0076] Multiple monitoring stations are set up around the wind turbine blades. LiDAR, temperature sensor and wind speed sensor are set up at each monitoring station to form a multi-station collaborative monitoring network through all monitoring stations.

[0077] The number of the monitoring stations is 3-5, which are distributed in a ring around the wind turbine blades. The angle between adjacent monitoring stations is 60°-120°, and the distance between each monitoring station and the wind turbine blade is 10-30 meters. The scanning frequency of the laser radar is 10-20Hz, and the scanning accuracy is better than 1mm.

[0078] The monitoring data of the wind turbine blades is collected through the multi-site collaborative monitoring network, and the monitoring data is transmitted using a wireless sensor network. The wireless sensor network includes multiple wireless sensor nodes and a data aggregation node. Each monitoring site corresponds to a wireless sensor node. The wireless sensor node adopts an adaptive sampling frequency mechanism to dynamically adjust the sampling frequency of the monitoring data according to the change rate of the monitoring data.

[0079] Specifically, this embodiment achieves all-round monitoring of wind turbine blades without blind spots through a multi-site ring-distributed monitoring network layout, avoids monitoring blind spots, and improves the comprehensiveness of monitoring. The accuracy of monitoring data is ensured by reasonably setting the angle (60°~120°) and distance (10-30 meters) of the monitoring sites, as well as high-performance lidar (10-20Hz scanning frequency, accuracy better than 1mm). Wireless sensor networks are used for data transmission, which improves the efficiency and reliability of data transmission and reduces wiring costs and maintenance difficulties.

[0080] The wireless sensor nodes adopt an adaptive sampling frequency mechanism, which can dynamically adjust the sampling frequency according to the change rate of the monitoring data, optimizing the utilization of system resources while ensuring the monitoring effect and improving the monitoring efficiency.

[0081] The environmental factor compensation model based on the deep learning network is constructed, specifically including:

[0082] A dual-branch deep neural network model including a temperature compensation subnetwork and a wind compensation subnetwork is constructed, wherein the temperature compensation subnetwork includes a feature extraction layer, a physical constraint layer, and a compensation coefficient generation layer; the feature extraction layer adopts a multi-layer perceptron structure to extract the spatiotemporal characteristics of temperature data; the physical constraint layer is used to calculate the temperature strain coefficient; the compensation coefficient generation layer generates a temperature compensation coefficient matrix of three-dimensional point cloud data based on the temperature strain coefficient;

[0083] The calculation formula of the temperature strain coefficient is:

[0084] ε1=α1×ΔT1×(1+β1×sin(ω1t1))×exp(-γ1×d1);

[0085] Among them, ε1 is the temperature strain coefficient, α1 is the linear thermal expansion coefficient of the fan blade material, ΔT1 is the temperature change, β1 is the temperature nonlinear correction coefficient, ω1 is the temperature cycle change frequency, t1 is the first time parameter, exp(·) is the exponential function, γ1 is the temperature spatial attenuation coefficient, and d1 is the distance from the measurement point to the blade surface;

[0086] The calculation formula of the temperature compensation coefficient matrix is:

[0087]

[0088]

[0089] Where M is the temperature compensation coefficient matrix, W is the spatial weight matrix, ⊙ is the Hadamard product, ε1 is the temperature strain coefficient, I is the unit matrix, σ1 is the temperature gradient coefficient, ▽T2 is the temperature gradient matrix, δ1 is the conduction attenuation coefficient, ρ is the spatial attenuation rate, R is the distance matrix, exp(·) is the exponential function, P is the current point cloud coordinate matrix, P ref is the reference point cloud coordinate matrix, τ1 is the weight adjustment parameter, and ||·||2 is the Euclidean distance norm;

[0090] The wind compensation subnetwork includes a feature extraction layer, a physical constraint layer and a compensation coefficient generation layer; the feature extraction layer adopts a long short-term memory network structure to extract the time series characteristics of wind speed data; the physical constraint layer introduces the aerodynamic characteristics of the blade to calculate the wind compensation coefficient; the compensation coefficient generation layer generates a wind compensation coefficient matrix of point cloud data based on the wind compensation coefficient;

[0091] The calculation formula of the wind compensation coefficient is:

[0092]

[0093] Among them, η1 is the wind compensation coefficient, k1 is the wind coefficient, v is the real-time wind speed, v0 is the reference wind speed, μ1 is the direction correction coefficient, θ is the wind direction angle, λ1 is the height attenuation coefficient, and h1 is the relative height;

[0094] Pre-training the dual-branch deep neural network through simulation data to obtain an initial training environmental factor compensation model, acquiring historical monitoring data, fine-tuning the initial training environmental factor compensation model based on the historical monitoring data to obtain an environmental factor compensation model to be optimized, and continuously optimizing the model parameters of the environmental factor compensation model to be optimized by online learning to obtain an environmental factor compensation model; adopting an adaptive learning rate adjustment mechanism during the training process to dynamically adjust the learning rate of the model according to the performance of the validation set;

[0095] The calculation formula of the environmental factor compensation model is:

[0096]

[0097] Among them, P comp is the compensated point cloud data, P orig is the input original point cloud data, ⊙ is the Hadamard product, M is the temperature compensation coefficient matrix, N is the wind compensation coefficient matrix, is the time-varying adaptive weight function, and C is the spatial constraint matrix.

[0098] Specifically, this embodiment constructs a dual-branch deep neural network model to compensate for the temperature and wind effects respectively, thereby achieving accurate modeling and compensation for different environmental factors, improving the accuracy of point cloud data, introducing a physical constraint layer and an accurate mathematical model in the compensation subnetwork, including a temperature strain coefficient and a compensation coefficient matrix, combining physical laws with deep learning, and making the compensation results more in line with actual conditions. The wind compensation subnetwork uses a long short-term memory network model structure to extract time series features, which can better capture the time dependence of wind speed data and improve the modeling accuracy of wind effects. This embodiment uses an adaptive learning rate adjustment mechanism and an online learning method to enable the model to continuously optimize and adapt to environmental changes, thereby improving robustness.

[0099] The environmental factor compensation model is used to perform wind compensation and temperature compensation on the three-dimensional point cloud data according to the temperature data and the wind speed data to obtain compensated point cloud data, specifically including:

[0100] A sliding time window mechanism is used to process temperature data and wind speed data, and the window length is adaptively adjusted according to the data change rate of the temperature data and wind speed data; feature extraction and normalization are performed on the temperature data and wind speed data in each time window to generate a feature vector; the feature vector is input into the dual-branch network of the environmental factor compensation model to obtain the output of the temperature compensation branch and the wind compensation branch respectively; the weight coefficients of the two branches are calculated through the attention mechanism, and they are fused based on the adaptive weight function to obtain a comprehensive compensation coefficient matrix;

[0101] An iterative optimization algorithm based on local consistency constraints is used for point cloud compensation, and each iteration includes: calculating the residual between the current point cloud data and the local reference model; updating the comprehensive compensation coefficient matrix based on the residual; correcting the current point cloud data using the updated comprehensive compensation coefficient matrix; calculating the local curvature change of the corrected point cloud data; terminating the iterative process when the curvature change of two consecutive iterations is less than a preset threshold; the local consistency constraint takes into account the geometric characteristics and topological relationships of the point cloud to ensure the spatial continuity of the compensation result.

[0102] Specifically, this embodiment improves the flexibility and adaptability of temperature and wind speed data processing by adopting a sliding time window mechanism and adaptively adjusting the window length according to the data change rate, and can better adapt to the data characteristics under different working conditions. The weight coefficients of the two branches of temperature compensation and wind compensation are calculated using the attention mechanism, and are fused through an adaptive weight function to achieve accurate modeling and dynamic balance of the influence of different environmental factors, thereby improving the accuracy of compensation. Point cloud compensation is performed using an iterative optimization algorithm based on local consistency constraints. By considering the geometric features and topological relationships of the point cloud, the spatial continuity of the compensation results is ensured, and unreasonable deformation during the compensation process is avoided. This embodiment achieves automatic convergence of the compensation process and ensures the reliability of the compensation results through mechanisms such as residual calculation, coefficient update, and local curvature change judgment.

[0103] The optimization processing of the compensated point cloud data includes noise filtering, data alignment and dynamic updating to obtain optimized point cloud data, specifically including:

[0104] Noise filtering is performed on the compensated point cloud data, including: calculating the local neighborhood features of each point, the local neighborhood features including point density, curvature and normal vector, setting a density threshold and a curvature threshold based on statistical feature analysis, removing low-density points and high-curvature abnormal points based on the density threshold and the curvature threshold, and performing probability filtering based on Gaussian distribution on the remaining point cloud data to further remove isolated points and noise points;

[0105] The calculation formula of the point density is:

[0106]

[0107] Among them, ρ i For point p i The local density of k2 is the number of neighborhood points, ||p i -p j || is point p i and the neighborhood point p j The Euclidean distance, σ2 is the density adjustment parameter;

[0108] Perform data alignment and dynamic update on the noise-filtered point cloud data, including: using an alignment method based on an iterative closest point algorithm to calculate the rigid body transformation matrix between the current point cloud data and the reference model, using a Kalman filter algorithm to dynamically update the point cloud data based on real-time monitoring data, correcting the time series error of the point cloud, and obtaining optimized point cloud data. In the dynamic update process, the local geometric features of the point cloud are combined to ensure the spatial continuity and accuracy of the alignment result;

[0109] The calculation formula of the rigid body transformation matrix is:

[0110]

[0111] Where T is the rigid body transformation matrix, p m is the mth point of the current point cloud data, q m is the corresponding m-th point in the reference model, k3 is the total number of points, is the rigid body transformation matrix T when the function value is the smallest;

[0112] The calculation formula for the state update of the Kalman filter algorithm is:

[0113] x tt =x tt-1 +K t (z t -Hx tt-1 );

[0114] Among them, x tt is the estimated value of the current state, x tt-1 is the predicted value of the previous moment, K t is the Kalman gain coefficient, z t is the current observation value, and H is the observation matrix.

[0115] Specifically, this embodiment removes abnormal points, isolated points and noise points in the point cloud data through a multi-level noise filtering mechanism based on local neighborhood features, including point density, curvature, and normal vectors, combined with statistical feature analysis and Gaussian distribution probability filtering, thereby improving the reliability of the point cloud data. An alignment method based on an iterative nearest point algorithm is adopted to achieve precise alignment of the point cloud data with the reference model by calculating the rigid body transformation matrix, thereby improving the accuracy of data registration. The Kalman filter algorithm is used to dynamically update the point cloud data, correct the time series error of the point cloud, and improve the time series continuity of the data. In the process of data processing, the local geometric features of the point cloud are combined to ensure the spatial continuity of the alignment results, avoid unreasonable local deformation, and improve the overall optimization effect.

[0116] The constructing of a three-dimensional fan blade model based on the optimized point cloud data, and obtaining the fan blade deformation data by comparing the three-dimensional fan blade model with the blade design model, specifically includes:

[0117] A point cloud fitting method based on B-spline surface is used to extract key feature points of optimized point cloud data, and surface reconstruction is performed based on the key feature points to build a three-dimensional fan blade model. During the surface reconstruction process, the geometric characteristics of the blade and boundary constraints are combined to ensure the smoothness and accuracy of the model.

[0118] The calculation formula for the surface reconstruction of the three-dimensional fan blade model is:

[0119]

[0120] Among them, S(u,w) is the surface reconstructed in the u and w directions, N a,p(u) is the first B-spline basis function, N b,q(w) is the second B-spline basis function, P a,b is the control point, k4 is the number of control points in the u direction minus 1, k5 is the number of control points in the w direction minus 1, p is the order of the first B-spline, and q is the order of the second B-spline;

[0121] A comparative analysis method based on point cloud distance is used to align the wind turbine 3D blade model with the blade design model, calculate the deformation of each point, and generate a deformation thermal map of the blade based on the deformation distribution. When calculating the deformation, the local curvature and normal vector of the blade are combined to ensure the accuracy of the deformation data.

[0122] The calculation formula of the deformation is:

[0123] Δd c =||p c -q c ||+α2·|κ c -κ ref |;

[0124] Where, Δd c is the cth point p c The deformation, q c is the position of the corresponding c-th point in the design model, κ c and κ ref are the local curvature functions of the current model and the design model respectively, and α2 is the curvature weight coefficient;

[0125] The calculation formula of the deformation thermodynamic map is:

[0126]

[0127] Where H(e,f) is the deformation value of the point (e,f) in the thermal diagram, Δd g For point p g The deformation amount, σ3 is the spatial scale parameter, and k6 is the number of points in the deformation thermodynamic map.

[0128] Specifically, this embodiment achieves high-precision reconstruction of the three-dimensional model of the blade by adopting a point cloud fitting method based on B-spline surfaces, combined with the geometric characteristics and boundary constraints of the blade, and ensures the smoothness and geometric accuracy of the model. By extracting key feature points for surface reconstruction, the computational complexity is reduced, the efficiency of model construction is improved, and the retention of important detail features is ensured. By adopting a comparative analysis method based on point cloud distance, combined with the local curvature and normal vector information of the blade, high-precision deformation calculation is achieved, and the accuracy of deformation monitoring is improved. By generating a deformation heat map, the deformation distribution of each area of ​​the blade is intuitively displayed, which is convenient for staff to quickly identify and locate the deformation area, and improves the visualization effect of the monitoring results.

[0129] The real-time analysis of the fan blade deformation data and outputting a deformation warning signal when the fan blade deformation exceeds a preset threshold value specifically includes:

[0130] Performing multi-dimensional analysis on the deformation data of the fan blades to obtain blade deformation characteristic parameters, wherein the multi-dimensional analysis includes deformation distribution analysis, deformation trend analysis and deformation rate analysis;

[0131] The deformation distribution analysis includes calculating the maximum deformation, average deformation and deformation standard deviation of each area of ​​the blade. The deformation trend analysis includes calculating the deformation change rate at multiple consecutive time points and performing trend fitting. The deformation rate analysis includes calculating the instantaneous deformation rate and cumulative deformation of each key point of the blade. The above analysis results are comprehensively evaluated to obtain a characteristic parameter set reflecting the overall deformation state of the blade.

[0132] Perform early warning judgment based on the blade deformation characteristic parameters, and when any characteristic parameter of the blade deformation characteristic parameters exceeds the corresponding preset threshold, determine the early warning level and output a corresponding deformation early warning signal;

[0133] Three warning levels are set, including mild warning, moderate warning and severe warning. Mild warning corresponds to local deformation exceeding the first threshold or deformation rate exceeding the first rate threshold, moderate warning corresponds to overall deformation exceeding the second threshold or deformation acceleration exceeding the acceleration threshold, severe warning corresponds to deformation of any monitoring point exceeding the third threshold or sudden deformation occurs. Different warning levels output different forms of warning signals, including system prompts, sound and light alarms and emergency stop commands.

[0134] Specifically, this embodiment achieves a comprehensive assessment of the deformation state of the blades by performing a multi-dimensional analysis of the deformation distribution, deformation trend and deformation rate of the deformation data of the wind turbine blades. A complete deformation state assessment system is constructed by using a set of characteristic parameters including the maximum deformation, average deformation and deformation standard deviation, combined with deformation trend fitting and rate calculation. Three warning levels are set, namely, mild warning, moderate warning and severe warning, and corresponding warning signals (system prompts, sound and light alarms, emergency stop instructions) are configured for different levels, thus achieving differentiated and refined management of the warning mechanism and improving the pertinence and effectiveness of the warning. By comprehensively considering multiple parameters such as local deformation, overall deformation, deformation rate, deformation acceleration, etc., a multi-dimensional warning judgment mechanism is established, which improves the accuracy and reliability of warning decisions and effectively reduces the risks of false alarms and missed alarms.

[0135] See also Figure 2 The present invention also provides a fan blade deformation monitoring system based on a three-dimensional fast point cloud model, the system comprising:

[0136] A data acquisition module, used to collect monitoring data of wind turbine blades through multiple types of sensors, wherein the monitoring data includes three-dimensional point cloud data, temperature data and wind speed data;

[0137] Compensation model module, used to build an environmental factor compensation model based on a deep learning network;

[0138] A point cloud compensation module, used to perform wind force compensation and temperature compensation on the three-dimensional point cloud data according to the temperature data and the wind speed data through the environmental factor compensation model to obtain compensated point cloud data;

[0139] A point cloud optimization module, used for optimizing the compensated point cloud data, wherein the optimization includes noise filtering, data alignment and dynamic updating, to obtain optimized point cloud data;

[0140] A deformation comparison module is used to construct a three-dimensional fan blade model based on the optimized point cloud data, and obtain fan blade deformation data by comparing the three-dimensional fan blade model with a blade design model;

[0141] The deformation warning module is used to analyze the deformation data of the fan blades in real time and output a deformation warning signal when the deformation amount of the fan blades exceeds a preset threshold.

[0142] Specifically, a fan blade deformation monitoring system based on a three-dimensional fast point cloud model in this embodiment constructs a fan blade deformation monitoring system by integrating a data acquisition module, a compensation model module, a point cloud compensation module, an optimization processing module, a three-dimensional model construction module, and a deformation warning module. The system can efficiently collect three-dimensional point cloud data, temperature data, and wind speed data of fan blades, use the environmental factor compensation model constructed by the deep learning network to compensate for wind force and temperature on the point cloud data, optimize the quality of the point cloud data, and construct a high-precision three-dimensional blade model based on the optimized point cloud data. By analyzing the blade deformation data in real time and outputting a graded warning signal, the system realizes accurate monitoring, dynamic evaluation, and timely warning of fan blade deformation, providing reliable protection for the safe operation of the fan.

[0143] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory communication interface and a bus: wherein the processor, memory and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement a fan blade deformation monitoring method based on a three-dimensional fast point cloud model.

[0144] The present invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, wherein the computer instructions enable the computer to implement all or part of the steps of a method for monitoring deformation of a wind turbine blade based on a three-dimensional fast point cloud model according to an embodiment of the present invention. The storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0145] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for monitoring fan blade deformation based on a three-dimensional fast point cloud model, characterized in that: The following steps are involved: Acquire monitoring data of wind turbine blades through multiple types of sensors, wherein the monitoring data includes three-dimensional point cloud data, temperature data and wind speed data; Constructing an environmental factor compensation model based on a deep learning network; By using the environmental factor compensation model, wind force compensation and temperature compensation are performed on the three-dimensional point cloud data according to the temperature data and the wind speed data to obtain compensated point cloud data; Performing optimization processing on the compensated point cloud data, wherein the optimization processing includes noise filtering, data alignment and dynamic updating to obtain optimized point cloud data; Building a three-dimensional fan blade model based on the optimized point cloud data, and obtaining fan blade deformation data by comparing the three-dimensional fan blade model with a blade design model; The fan blade deformation data is analyzed in real time, and when the deformation amount of the fan blade exceeds a preset threshold, a deformation warning signal is output.

2. A method for monitoring deformation of wind turbine blades based on a three-dimensional fast point cloud model as claimed in claim 1, characterized in that: The monitoring data of the wind turbine blades is obtained by using multiple types of sensors, and the monitoring data includes three-dimensional point cloud data, temperature data and wind speed data, specifically including: Multiple monitoring stations are set up around the wind turbine blades. LiDAR, temperature sensor and wind speed sensor are set up at each monitoring station to form a multi-station collaborative monitoring network through all monitoring stations. The monitoring data of the wind turbine blades is collected through the multi-site collaborative monitoring network, and the monitoring data is transmitted using a wireless sensor network. The wireless sensor network includes multiple wireless sensor nodes and a data aggregation node. Each monitoring site corresponds to a wireless sensor node. The wireless sensor node adopts an adaptive sampling frequency mechanism to dynamically adjust the sampling frequency of the monitoring data according to the change rate of the monitoring data.

3. A method for monitoring fan blade deformation based on a three-dimensional fast point cloud model as claimed in claim 2, characterized in that: The environmental factor compensation model based on the deep learning network is constructed, specifically including: A dual-branch deep neural network model including a temperature compensation subnetwork and a wind compensation subnetwork is constructed, wherein the temperature compensation subnetwork includes a feature extraction layer, a physical constraint layer, and a compensation coefficient generation layer; the feature extraction layer adopts a multi-layer perceptron structure to extract the spatiotemporal characteristics of temperature data; the physical constraint layer is used to calculate the temperature strain coefficient; the compensation coefficient generation layer generates a temperature compensation coefficient matrix of three-dimensional point cloud data based on the temperature strain coefficient; The calculation formula of the temperature strain coefficient is: ε1=α1×ΔT1×(1+β1×sin(ω1t1))×exp(-γ1×d1); Among them, ε1 is the temperature strain coefficient, α1 is the linear thermal expansion coefficient of the fan blade material, ΔT1 is the temperature change, β1 is the temperature nonlinear correction coefficient, ω1 is the temperature cycle change frequency, t1 is the first time parameter, exp(·) is the exponential function, γ1 is the temperature spatial attenuation coefficient, and d1 is the distance from the measurement point to the blade surface; The calculation formula of the temperature compensation coefficient matrix is: M=W⊙[ε1·I+σ1·▽T2+δ1·exp(-ρ·R)]; Where M is the temperature compensation coefficient matrix, W is the spatial weight matrix, ⊙ is the Hadamard product, ε1 is the temperature strain coefficient, I is the unit matrix, σ1 is the temperature gradient coefficient, ▽T2 is the temperature gradient matrix, δ1 is the conduction attenuation coefficient, ρ is the spatial attenuation rate, R is the distance matrix, exp(·) is the exponential function, P is the current point cloud coordinate matrix, P ref is the reference point cloud coordinate matrix, τ1 is the weight adjustment parameter, and ||·||2 is the Euclidean distance norm; Pre-training the dual-branch deep neural network through simulation data to obtain an initial training environmental factor compensation model, acquiring historical monitoring data, fine-tuning the initial training environmental factor compensation model based on the historical monitoring data to obtain an environmental factor compensation model to be optimized, and continuously optimizing the model parameters of the environmental factor compensation model to be optimized by online learning to obtain an environmental factor compensation model; adopting an adaptive learning rate adjustment mechanism during the training process to dynamically adjust the learning rate of the model according to the performance of the validation set; The calculation formula of the environmental factor compensation model is: Among them, P comp is the compensated point cloud data, P orig is the input original point cloud data, ⊙ is the Hadamard product, M is the temperature compensation coefficient matrix, N is the wind compensation coefficient matrix, is the time-varying adaptive weight function, and C is the spatial constraint matrix.

4. The method for monitoring fan blade deformation based on a three-dimensional fast point cloud model according to claim 1, characterized in that: The environmental factor compensation model is used to perform wind compensation and temperature compensation on the three-dimensional point cloud data according to the temperature data and the wind speed data to obtain compensated point cloud data, specifically including: A sliding time window mechanism is used to process temperature data and wind speed data, and the window length is adaptively adjusted according to the data change rate of the temperature data and wind speed data; feature extraction and normalization are performed on the temperature data and wind speed data in each time window to generate a feature vector; the feature vector is input into the dual-branch network of the environmental factor compensation model to obtain the output of the temperature compensation branch and the wind compensation branch respectively; the weight coefficients of the two branches are calculated through the attention mechanism, and they are fused based on the adaptive weight function to obtain a comprehensive compensation coefficient matrix; An iterative optimization algorithm based on local consistency constraints is used for point cloud compensation. Each iteration includes: calculating the residual between the current point cloud data and the local reference model; updating the comprehensive compensation coefficient matrix based on the residual; correcting the current point cloud data using the updated comprehensive compensation coefficient matrix; calculating the local curvature change of the corrected point cloud data; and terminating the iterative process when the curvature change of two consecutive iterations is less than a preset threshold.

5. The method for monitoring deformation of wind turbine blades based on a three-dimensional fast point cloud model according to claim 1, characterized in that: The optimization processing of the compensated point cloud data includes noise filtering, data alignment and dynamic updating to obtain optimized point cloud data, specifically including: Noise filtering is performed on the compensated point cloud data, including: calculating the local neighborhood features of each point, the local neighborhood features including point density, curvature and normal vector, setting a density threshold and a curvature threshold based on statistical feature analysis, removing low-density points and high-curvature abnormal points based on the density threshold and the curvature threshold, and performing probability filtering based on Gaussian distribution on the remaining point cloud data; The point cloud data after noise filtering is aligned and dynamically updated, including: using an alignment method based on an iterative closest point algorithm to calculate the rigid body transformation matrix between the current point cloud data and the reference model, and based on real-time monitoring data, using a Kalman filter algorithm to dynamically update the point cloud data, correcting the time series error of the point cloud, and obtaining optimized point cloud data.

6. A method for monitoring fan blade deformation based on a three-dimensional fast point cloud model as claimed in claim 5, characterized in that: The constructing of a three-dimensional fan blade model based on the optimized point cloud data, and obtaining the fan blade deformation data by comparing the three-dimensional fan blade model with the blade design model, specifically includes: A point cloud fitting method based on B-spline surface is used to extract key feature points of optimized point cloud data, and surface reconstruction is performed based on the key feature points to construct a three-dimensional blade model of the wind turbine; The calculation formula for the surface reconstruction of the three-dimensional fan blade model is: Among them, S(u,w) is the surface reconstructed in the u and w directions, N a,p(u) is the first B-spline basis function, N b,q(w) is the second B-spline basis function, P a,b is the control point, k4 is the number of control points in the u direction minus 1, k5 is the number of control points in the w direction minus 1, p is the order of the first B-spline, and q is the order of the second B-spline; A comparative analysis method based on point cloud distance is used to align the wind turbine 3D blade model with the blade design model, calculate the deformation of each point, and generate a deformation heat map of the blade based on the deformation distribution. The calculation formula of the deformation is: Δd c =||p c -q c ||+α2·|κ c -k ref |; Where, Δd c is the cth point p c The deformation, q c is the position of the corresponding c-th point in the design model, κ c and κ ref are the local curvature functions of the current model and the design model respectively, and α2 is the curvature weight coefficient; The calculation formula of the deformation thermodynamic map is: Where H(e,f) is the deformation value of the point (e,f) in the thermal diagram, Δd g For point p g The deformation amount, σ3 is the spatial scale parameter, and k6 is the number of points in the deformation thermodynamic map.

7. The method for monitoring fan blade deformation based on a three-dimensional fast point cloud model according to claim 1, characterized in that: The real-time analysis of the fan blade deformation data and outputting a deformation warning signal when the fan blade deformation exceeds a preset threshold value specifically includes: Performing multi-dimensional analysis on the deformation data of the fan blades to obtain blade deformation characteristic parameters, wherein the multi-dimensional analysis includes deformation distribution analysis, deformation trend analysis and deformation rate analysis; An early warning judgment is performed based on the blade deformation characteristic parameters. When any characteristic parameter of the blade deformation characteristic parameters exceeds a corresponding preset threshold, an early warning level is determined and a corresponding deformation early warning signal is output.

8. A fan blade deformation monitoring system based on a three-dimensional fast point cloud model, characterized in that: The system comprises: A data acquisition module, used to collect monitoring data of wind turbine blades through multiple types of sensors, wherein the monitoring data includes three-dimensional point cloud data, temperature data and wind speed data; Compensation model module, used to build an environmental factor compensation model based on a deep learning network; A point cloud compensation module, used to perform wind force compensation and temperature compensation on the three-dimensional point cloud data according to the temperature data and the wind speed data through the environmental factor compensation model to obtain compensated point cloud data; A point cloud optimization module, used for optimizing the compensated point cloud data, wherein the optimization includes noise filtering, data alignment and dynamic updating, to obtain optimized point cloud data; A deformation comparison module is used to construct a three-dimensional fan blade model based on the optimized point cloud data, and obtain fan blade deformation data by comparing the three-dimensional fan blade model with a blade design model; The deformation warning module is used to analyze the deformation data of the fan blades in real time and output a deformation warning signal when the deformation amount of the fan blades exceeds a preset threshold.

9. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other via the bus, the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

Images