Rotating blade video data synthesis method based on rigid-flexible coupling visual mapping and physical fidelity

Abstract

This invention relates to a method for synthesizing video data of rotating blades based on rigid-flexible coupling visual mapping and physical fidelity. The method includes: modal synthesis and reduction of flexible support components and exporting modal neutral files; multi-body topology assembly with an equivalent rigid body matrix model; and construction of a rigid-flexible coupling benchmark dynamic model of the entire machine using bushing force elements. Under non-stationary operating conditions, physical faults such as structural loosening and mass eccentricity are parametrically injected, and three-dimensional high-frequency displacement data of the observation points are solved and exported. Two-dimensional pixel displacement trajectories are generated based on virtual camera perspective projection mapping while fully preserving high-frequency micro-vibration characteristics. Sensor thermal noise and low-frequency large-amplitude drift signals are superimposed, and fault combinations and operating conditions are iteratively traversed within the parameter space to synthesize a video stream dataset. This invention establishes a quantitative cross-scale evolutionary mapping relationship from microscopic physical damage to macroscopic pixel micro-vibration trajectories, providing a high-fidelity data source for predictive maintenance and visual diagnostic algorithms for rotating blade equipment.

Description

Technical Field

[0001] This invention relates to the fields of machine vision and computer technology, and in particular to a method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity. Background Technology

[0002] Rotating blade equipment generally faces the dynamic problem of high-frequency vibration of flexible components coupled with rigid substrate. In actual operation, the rotating working parts and flexible support components of such equipment are subjected to strong coupling effects of high-speed rotation and unsteady aerodynamic loads for a long time, which makes them prone to early physical damage failures such as microcracks, mass eccentricity and loose connections.

[0003] In recent years, in particular, with the rapid development of sensor technology and flight control algorithms, electric vertical take-off and landing (eVTOL) aircraft, as the core carrier of the low-altitude economy, have become the most representative and demanding application of rotating blade equipment. In densely populated urban low-altitude airspace with complex airflow environments, the rotor, as the sole actuator, is constantly subjected to high-speed rotation and complex, ever-changing environments. Therefore, accurate early monitoring of minute damage to such rotating blade equipment is crucial for preventing major safety accidents.

[0004] The following detailed explanation uses eVTOL as a typical embodiment, but the aforementioned technical challenges also exist in rotating blade equipment such as aero engines, steam turbines, and wind turbines. However, existing structural health monitoring technologies currently have the following limitations when dealing with such rigid-flexible coupled rotating machinery: (1) Traditional monitoring based on contact or airborne sensing is limited by environmental noise and sampling difficulties.

[0005] Traditional contact monitoring methods, such as accelerometers, acoustic sensors, or strain gauges, are extremely difficult to deploy and highly invasive. This not only alters the inherent mechanical properties of the structure and significantly increases the load on the equipment, but also makes the signals highly susceptible to complex environmental noise, making it difficult to effectively isolate early, weak structural fault characteristics from background noise in low signal-to-noise ratio environments. While using airborne built-in sensors for diagnosis eliminates the need for additional loads, significant technical bottlenecks exist: firstly, limited by airborne bus bandwidth and data processing capabilities, the sampling frequency of conventional sensors is far lower than the high-speed rotational characteristic frequency of the rotating blades, leading to severe data aliasing and loss of high-frequency transient fault information; secondly, weak fault signals are easily overwhelmed by the dynamic response of the flight attitude control closed loop, resulting in a lack of significant identifiability in both the time and frequency domains, making it difficult to accurately capture millisecond-level transient characteristics.

[0006] (2) Non-contact diagnosis based on computer vision lacks physical mechanism support and high-fidelity data source.

[0007] Non-contact vibration measurement and health monitoring methods based on computer vision theoretically have broad application prospects. However, due to the extreme scarcity of real physical fault samples, existing visual diagnostic techniques still have significant shortcomings: First, although event cameras have the advantage of asynchronous triggering, their raw output data is difficult to directly interface with general deep learning architectures and often needs to be reconstructed into frame-like images through spatiotemporal aggregation. This process introduces irreversible reconstruction artifacts and spatiotemporal information aliasing, resulting in the loss of the ability to finely represent the high-frequency micro-vibration characteristics of rotating parts. Second, although micro-motion video magnification technology can achieve sub-pixel-level displacement enhancement, such methods are extremely dependent on high-quality visual perception data. Currently, the industry severely lacks high-fidelity physical video datasets that can cover dynamic evolution under multiple operating conditions such as takeoff or variable speed and accurately represent the structural fault mechanism, leading to a long-term data shortage for the training of diagnostic models. Third, most of the current mainstream virtual data synthesis and visual diagnostic algorithms are based on pure data-driven approaches, completely detached from the physical process from structural damage evolution to dynamic response and then to pixel displacement mapping. This black-box modeling approach, lacking physical mechanism support, cannot truly reflect the characteristics of rigid-flexible coupling. As a result, the model has extremely poor generalization ability when faced with uncovered operating conditions or airflow disturbances, and the diagnostic results lack physical interpretability, making it difficult to meet the stringent requirements of aviation airworthiness certification and predictive maintenance for core industrial equipment.

[0008] Chinese patent application publication number CN121757391A discloses a virtual-real fusion test method for unmanned aerial vehicles (UAVs) based on sensor injection and digital twins. This method constructs a virtual 3D scene and generates a corresponding digital twin model through sensor injection and digital twin fusion, enabling the injection of virtual sensor signals and closed-loop control. Combined with deep fusion of virtual obstacles and physical entities, it forms a cross-temporal and spatial virtual-real feedback closed-loop test link. However, its simulation fidelity only reaches the electrical signal level and does not achieve visual-physical fidelity. This method cannot provide a synthetic video stream containing local high-frequency deformation features.

[0009] In summary, there is currently a lack of a method for synthesizing rotating blade video data to solve or partially solve the aforementioned problems. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a method for synthesizing video data of rotating blades based on rigid-flexible coupling visual mapping and physical fidelity, so as to solve or partially solve the problems of lack of physical damage mechanism support, low physical fidelity, and inability to quantitatively traverse fault features under complex working conditions in the visual diagnosis of rotating blades.

[0011] The objective of this invention can be achieved through the following technical solutions: One aspect of the present invention provides a method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity, comprising the following steps: S1, based on the rigid base model of the rotating blade equipment, performs rigid body geometry simplification and physical property equivalence to obtain the equivalent rigid base model and its physical property parameters, and performs free interface mode reduction for the flexible support / connection components in the vibration transmission path to extract multi-order regular modes. S2, based on the equivalent rigid body matrix model and the multi-order regular modes, multi-body topology assembly is performed to establish a nonlinear contact mechanics model between the supporting component and the supporting ground. In this model, a bushing force element is used to simulate the connection structure between the flexible component and the rigid matrix, and a stiffness matrix representing the initial connection state is assigned. The stiffness matrix is ​​dynamically adjusted in the parameter space to simulate the degradation process of the structural connection state. Observation markers are configured on the surface of the observable area of ​​the matrix to extract and output the dynamic response of the high-frequency micro-vibration of the structure, and a rigid-flexible coupling reference dynamic model of the whole machine is constructed. S3, set non-stationary working condition boundaries in the benchmark dynamic model to simulate the complex working conditions of the rotating blade equipment in actual operation, apply unsteady aerodynamic loads, inject physical damage faults into the system parameterization in the parameter space, solve the transient dynamics of operation, and calculate the original displacement time series data and acceleration time series data of the observation markers in three-dimensional space. S4. Construct a virtual camera intrinsic parameter matrix. Utilize the principle of three-dimensional coordinate perspective projection to establish a mapping operator from three-dimensional physical space to a two-dimensional image plane. The original displacement time series data is perspective-projected and mapped to the original pixel displacement trajectory on the two-dimensional image plane. The mapping process completely preserves the high-frequency micro-vibration characteristics in the original displacement. S5, environmental interference noise is superimposed on the original pixel displacement trajectory, and a low-frequency drift signal is injected to simulate environmental gust disturbance and observation array sway interference, generating a virtual pixel trajectory dataset containing micro-vibration characteristics and environmental interference. By iteratively traversing different fault severity coefficients, multi-source physical damage fault combinations and non-stationary working conditions in the parameter space, a video stream dataset is synthesized to establish a cross-scale evolution mapping relationship from macroscopic physical damage of equipment to microscopic pixel micro-vibration trajectory.

[0012] As a preferred technical solution, the process of obtaining the equivalent rigid body matrix model and physical property parameters in step S1 includes the following steps: Remove internal non-load-bearing components and convert the total mass of electronic components into central counterweight parameters. Recalibrate and balance the total mass and center of gravity to obtain an equivalent rigid body matrix model and its physical property parameters. The process of extracting multi-order regular modes includes the following steps: For the flexible support / connection components in the vibration transmission path, a finite element mesh model is established, the orthogonal anisotropic physical parameters of the composite material are input, and the Craig-Bampton modal synthesis method is used to perform free interface mode reduction on the flexible component, and the first 15 to 20 canonical modes are extracted as the multi-order canonical modes.

[0013] As a preferred technical solution, in step S2, the nonlinear contact mechanics model is characterized by a bushing force element and a nonlinear contact model; wherein, the bushing force element includes a bushing force element used to simulate the connection state between the rotating blade and the motor shaft or transmission shaft, the bushing force element is used to simulate the bolt connection structure between the flexible component and the rigid substrate, the supporting component is the landing gear, and the nonlinear contact mechanics model is configured with a nonlinear contact stiffness coefficient, a damping coefficient and a Coulomb friction coefficient.

[0014] As a preferred technical solution, step S3, which involves injecting physical damage faults into the system parameterization within the parameter space, includes the following steps: Lowering the diagonal coefficient value in the stiffness matrix of the bushing force element simulates the decrease in bolt preload and the degradation state of the connection. An eccentric mass block is added at a specified blade node of the rotating working part to generate a centrifugal excitation force proportional to the square of the rotational speed, which characterizes the asymmetric mass eccentricity.

[0015] As a preferred technical solution, in S3, the boundary of the non-stationary operating condition is a preset rotary blade sweep frequency acceleration speed curve, and the unsteady aerodynamic load is a dynamic load calculated based on a quadratic aerodynamic model that changes nonlinearly with the rotational speed.

[0016] As a preferred technical solution, in step S4, the virtual camera intrinsic parameter matrix includes focal length, principal point coordinates, and distortion coefficients, and the three-dimensional coordinate perspective projection principle is the central perspective projection principle.

[0017] As a preferred technical solution, in step S5, the environmental interference noise is Gaussian white noise, used to simulate the thermal noise of the camera sensor, and the power spectral density of the low-frequency drift signal follows a preset low-frequency disturbance model, which includes a low-altitude gust disturbance model and a ground observation array sway interference model.

[0018] As a preferred technical solution, the rotating blade equipment includes an electric vertical takeoff and landing aircraft, a steam turbine, a wind turbine, an aircraft engine, or an industrial fan.

[0019] As a preferred technical solution, in step S5, the iterative traversal includes the following steps: By changing the fault severity coefficient within the parameter space and combining different physical damage faults, the system traverses various operating conditions, including non-stationary speeds, and synthesizes a video stream dataset.

[0020] In another aspect, an electronic device is provided, including one or more processors, a memory, and one or more programs stored in the memory, said one or more programs including instructions for executing the aforementioned method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity.

[0021] Compared with the prior art, the present invention has at least one of the following beneficial effects: (1) Fully extract the dynamic response of high-frequency elastic micro-vibration of flexible support components and rotating working parts: Traditional multibody dynamics modeling relies only on rigid body assumptions and cannot describe the local high-frequency elastic deformation characteristics of flexible support components under unsteady excitation, resulting in the loss of key physical micro-vibration information in the simulation data. In this regard, the present invention establishes a finite element mesh model for flexible support / connection components in the vibration transmission path, inputs material physical parameters, uses the modal synthesis method to perform free interface modal reduction on the flexible components, extracts multi-order regular modes, imports the equivalent rigid body matrix model and modal neutral file into the multibody dynamics simulation software, and performs multibody topology assembly to construct the whole machine rigid-flexible coupling benchmark dynamic model, which breaks through the limitations of rigid body modeling. The constructed rigid-flexible coupling model can accurately capture and output the dynamic response of high-frequency elastic micro-vibration of flexible support components and rotating working parts, providing high-fidelity source data with rich physical features for subsequent visual mapping.

[0022] (2) Strong generalization ability for minor damage: Existing visual diagnosis relies on pure data-driven algorithms. The synthetic data source is disconnected from the actual physical damage evolution mechanism of the equipment. Due to experimental costs and risks, it is difficult to systematically traverse complex working conditions and multiple fault severity combinations, resulting in insufficient generalization ability of the model for minor damage. In this regard, the present invention injects physical damage faults into the system parameterization in the parameter space. It simulates structural loosening faults by reducing the stiffness coefficient matrix of the bushing force element at a specific connection part, and simulates the dynamic unbalanced excitation load of the rotor by adding a small eccentric mass block at a specified node of the rotating working part. By iteratively traversing different fault severity coefficients and multi-source physical damage faults in the parameter space, the model can achieve the desired generalization ability for minor damage. By combining and non-stationary operating conditions, a video stream dataset with strict physical interpretability is synthesized, thereby quantitatively establishing a cross-scale evolutionary mapping relationship from microscopic physical damage of equipment to macroscopic pixel micro-vibration trajectories. Through the above process, physical faults with clear mechanical connotations are dynamically injected in a parameterized manner, establishing a quantitative cross-scale evolutionary model from macroscopic physical damage faults to microscopic pixel micro-vibration trajectories. This fills the gap of pure data-driven algorithms lacking explanation of physical damage mechanisms. At the same time, through automated cyclic traversal of the parameter space, a dataset covering extreme operating conditions and the entire fault spectrum can be efficiently generated without conducting high-cost and high-risk destructive experiments, improving the generalization detection capability of subsequent diagnostic models for minor damage.

[0023] (3) High physical fidelity: This invention constructs a virtual camera intrinsic parameter matrix and uses the principle of three-dimensional coordinate perspective projection to establish a mapping operator from three-dimensional physical space to two-dimensional image plane. The original displacement time series data of the observation point in three-dimensional space is mapped to the original pixel displacement trajectory on the two-dimensional image plane through perspective projection. The mapping process completely preserves the high-frequency micro-vibration features in the original displacement time series data, ensuring that the physical feature signals caused by the local high-frequency elastic micro-vibration of the structure are not filtered out. The constructed perspective projection mapping operator serves as a lossless physical feature transmission bridge, ensuring that the high-frequency elastic micro-vibration features in three-dimensional space are completely and faithfully mapped to the two-dimensional pixel displacement trajectory. This ensures the integrity and interpretability of the physical feature signals in the final video data and provides a high-quality data foundation for subsequent fine visual feature extraction.

[0024] (4) Enhanced anti-interference capability: This invention superimposes environmental interference noise onto the acquired original pixel displacement trajectory, including superimposing Gaussian white noise to simulate the thermal noise of the camera sensor, and simultaneously injects low-frequency large-amplitude drift signals to simulate environmental gust disturbances and observation array swaying interference, generating a virtual pixel trajectory dataset containing micro-vibration features and environmental interference. By equivalent modeling and superposition of multi-source environmental interference noise, the synthesized video stream dataset closely approximates the visual measurement conditions in the real service environment, possessing high physical fidelity, and effectively improving the generalization ability and anti-interference capability of subsequent visual diagnostic algorithms under low signal-to-noise ratio and harsh conditions.

[0025] (5) Constructing a data-algorithm closed loop and cross-equipment universality: The data synthesis mechanism constructed in this invention provides a data source with strict physical mechanism constraints for subsequent downstream diagnostic algorithms such as micro-vibration visual measurement and structural vibration spatiotemporal coherence analysis. It establishes a digital twin closed-loop verification paradigm from forward rigid-flexible dynamic evolution deduction to inverse visual feature inversion diagnosis, gets rid of dependence on the geometric dimensions and topology of a single specific equipment, and can be seamlessly transferred to complex rotating blade equipment such as steam turbines, wind turbines, and aero engines. It provides universal methodological support for the study of the evolution law of weak damage characteristics in multiple industrial application scenarios and the low-cost construction of high signal-to-noise ratio fault databases. Attached Figure Description

[0026] Figure 1 This is a flowchart of the rotating blade video data synthesis method based on rigid-flexible coupling visual mapping and physical fidelity in the embodiment; Figure 2 This is a schematic diagram of the rigid-flexible coupling reference dynamic model of the eVTOL system in the embodiment. Figure 3 This is a schematic diagram illustrating the geometric principle of mapping three-dimensional spatial displacement to two-dimensional image pixels in the embodiment. Figure 4 This is a schematic diagram of the electronic device in the embodiment. Among them, 201 is the equivalent rigid base, 202 is the flexible support / connection component, 203 is the elastic connection unit, 204 is the rotating working part, 205 is the virtual displacement observation point, 301 is the three-dimensional spatial physical displacement vector, 302 is the virtual camera center, and 303 is the two-dimensional image pixel displacement vector. Detailed Implementation

[0027] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0028] Example 1 To address the aforementioned problems in existing technologies, this embodiment provides a method for synthesizing video data of rotating blades based on rigid-flexible coupling visual mapping and physical fidelity. This method aims to solve engineering challenges in the structural health monitoring of rotating blade equipment, such as eVTOL, including the lack of physical damage mechanism support in visual diagnostic algorithms, low physical fidelity of the synthesized data source, and difficulty in quantitatively traversing fault features under complex operating conditions. By constructing a rigid-flexible coupling visual mapping and physical fidelity data synthesis method for rotating blade equipment, an automated mapping from the evolution of physical damage to visual imaging features is achieved, providing a high-fidelity data source with strict physical interpretability for predictive maintenance of rotating blade equipment (especially eVTOL).

[0029] The video data synthesized by the method can be used as input to existing visual diagnostic algorithms to output the working status of the rotating blade equipment. Preferably, the working status includes a 0-1 variable indicating whether it is normal, and may also include a fault classification label.

[0030] Preferably, this method can form a data-algorithm closed loop with existing visual measurement methods and systems for stroboscopic phase-locked micro-vibrations of rotating blades. The former provides the latter with a high-fidelity training and verification data source, and the synergistic application of the two constructs a complete digital twin diagnostic architecture covering physical evolution simulation and fine feature extraction.

[0031] This method combines multibody dynamics evolution mechanisms with multidimensional geometric perspective projection, breaking the black-box limitations of traditional data-driven visual diagnostics. Those skilled in the art should understand that although the following embodiments primarily use electric vertical takeoff and landing (eVTOL) aircraft as a typical application scenario for detailed explanation, the dynamic modeling, virtual space physical feature extraction, and cross-scale visual optical flow mapping logic of this invention have universal mechanical and geometrical mathematical essences, and are equally applicable to other general rotating blade equipment with rotating working parts and flexible support features, such as steam turbines, wind turbines, aero engines, and industrial fans.

[0032] For details, see Figure 1 The method includes the following steps: Step S1: Rigid body equivalent and flexible component mode extraction.

[0033] Based on the original 3D CAD model of the eVTOL fuselage, rigid body geometry simplification and physical property equivalence were performed. While maintaining the overall macroscopic topology, internal non-load-bearing wiring harnesses, circuit boards, and connecting fasteners were removed. The total mass of the internal electronic components was converted into central counterweight parameters. The overall mass and center of gravity of the fuselage were recalibrated and balanced to obtain the desired results. Figure 2The equivalent rigid matrix 201 model and its physical properties, such as inertia tensor, are shown. Simultaneously, structural mechanics constraint analysis is performed on key structural components in the eVTOL vibration transmission path. A finite element mesh model of the flexible carbon fiber arm is established, with input orthogonal anisotropic physical parameters of the composite material. Under linear configuration and boundary constraints, the Craig-Bampton modal synthesis method is used to perform free interface modal reduction and order reduction flexibility processing on the carbon fiber arm, extracting the first 15 to 20 orders of structural dynamic characteristic modes, and exporting a modal neutral file (MNF) as the flexible support / connection component 202, thus achieving dimensionality reduction of the dynamic equations.

[0034] Step S2: Multibody topology assembly and observation point configuration.

[0035] The equivalent rigid base 201 model obtained in step S1 and the modal neutral file of the flexible support / connection component 202 are imported into the multibody dynamics simulation platform for multibody topology assembly. A bushing force element is used as the elastic connection element 203 to simulate the bolted connection structure between the arm and fuselage, and its translational and rotational degrees of freedom are assigned stiffness and damping matrices representing the initial healthy state, respectively. A nonlinear contact mechanics model between the landing gear and the supporting ground is established, configuring the nonlinear contact stiffness coefficient, damping coefficient, and Coulomb friction coefficient. Several virtual displacement observation points 205 are configured on the observable surface of the equivalent rigid base 201 or the rotating working component 204 to extract and output the dynamic response of the high-frequency elastic micro-vibration of the structure in real time during the solution process, ultimately completing the construction of a high-fidelity whole-aircraft rigid-flexible coupling reference dynamic model.

[0036] See Figure 2 This example uses an electric vertical takeoff and landing (eVTOL) aircraft as a typical embodiment of rotating blade-type equipment for a concrete demonstration. In this embodiment, the equivalent rigid base 201 corresponds to the fuselage of the eVTOL, and is geometrically simplified as a rigid body without elastic deformation to focus on the vibration characteristics of the core transmission and actuation components. The flexible support / connection component 202 corresponds to the carbon fiber flexible arm of the eVTOL, retaining finite element mesh details to extract and characterize high-frequency structural dynamic modes. The elastic connection unit 203 corresponds to the bushing force unit simulating the bolted connection between the arm and the fuselage; its stiffness matrix can be parameterized to simulate degradation or loosening faults. The rotating working component 204 corresponds to the rotor of the eVTOL, serving as the main source of excitation loads and the core target area for visual monitoring. The virtual displacement observation point 205 corresponds to three-dimensional dynamic markers configured on the observable surface of the base or rotor, used to extract and output real-time three-dimensional spatial displacement time-series data of the structure after excitation in transient dynamics solutions.

[0037] Step 3: Setting up non-stationary operating conditions and injecting physical faults.

[0038] In step 2, non-stationary operating condition boundaries are set in the baseline dynamic model, such as a preset rotating blade sweep-frequency acceleration speed curve, to simulate complex operating conditions in actual equipment operation, and unsteady aerodynamic loads are applied, such as dynamic loads calculated based on a quadratic aerodynamic model. Based on this, physical damage faults are parametrically injected into the system within the parameter space, specifically including: simulating structural loosening faults by reducing the stiffness coefficient matrix of the bushing force element at specific connection points, and simulating dynamic unbalanced excitation loads of the rotor by adding small eccentric mass blocks at specified nodes of the rotating working parts, characterizing asymmetric mass eccentricity faults. The transient dynamics are solved, and the original displacement and acceleration time-series data of the observation markers in three-dimensional space are calculated and exported in real time.

[0039] Specifically, in this embodiment, a continuous rotor sweep-frequency acceleration speed curve is input to simulate the variable speed conditions in actual eVTOL flight, and quadratic unsteady aerodynamic forces and drag loads that vary nonlinearly with rotational speed are applied. Based on this, structural physical damage faults are parametrically injected into the system within the dynamic parameter space, specifically including: (1) Structural loosening fault: By lowering the diagonal coefficient value in the stiffness matrix of the bushing force unit at a specific elastic connection unit 203, the preload of the bolt connection and the connection degradation state are simulated. (2) Rotor dynamic imbalance fault: Add a small eccentric mass block at the designated blade node of the rotating working part 204 to generate a centrifugal excitation force proportional to the square of the rotation speed, which characterizes the asymmetric mass eccentricity.

[0040] Run the transient dynamics solver to perform time-domain integration, calculate and export in real time the original displacement time series data and acceleration time series data of each virtual displacement observation point 205 in three-dimensional space.

[0041] Step S4, virtual visual perspective projection mapping.

[0042] Construct the intrinsic parameter matrix of the virtual camera model, including focal length, principal point coordinates, and distortion coefficients. Utilize the principle of 3D coordinate center perspective projection to establish... Figure 3 The diagram shows the mapping operator from the three-dimensional simulated physical space to the two-dimensional image plane. The three-dimensional physical displacement vector 301 of the virtual displacement observation point 205 in the world coordinate system, derived in step 3, is converted into a two-dimensional image pixel displacement vector 303 in the two-dimensional image plane pixel coordinate system through the perspective projection operator of the virtual camera center 302.

[0043] Throughout the entire process of visual mapping and coordinate transformation, the high-frequency features in the original pixel displacement trajectory generated after spatial displacement transformation are fully preserved, ensuring that the physical feature signals caused by local high-frequency elastic micro-vibrations of the structure are not filtered out, and establishing a quantitative cross-scale evolution relationship from physical structural damage to macroscopic image pixel features.

[0044] See Figure 3 This is a schematic diagram illustrating the geometric principle of mapping three-dimensional spatial displacement to two-dimensional image pixels. The three-dimensional physical displacement vector 301 represents the actual spatial displacement of the observation point on the structure surface from its initial position P to its physical evolution position P' after being excited in the world coordinate system (X, Y, Z). This vector is directly derived from the aforementioned rigid-flexible coupling dynamic model, such as... Figure 2 The high-frequency dynamic response data output in real time from the virtual displacement observation point 205 contains the true mechanical micro-vibration characteristics of the structure. The virtual camera center 302, i.e., the lens optical center O in the perspective projection mapping, combined with the set virtual camera focal length and intrinsic / extrinsic parameter matrices, constructs a mathematical mapping operator connecting the three-dimensional physical simulation space and the two-dimensional virtual visual plane. The two-dimensional image pixel displacement vector 303 in the pixel coordinate system (… u , v In the image plane, the three-dimensional physical displacement is represented by pixels. p Move to p The visual displacement traces are obtained. This mapping process strictly follows the central perspective projection geometry mechanism, ensuring that the generated two-dimensional pixel displacements retain the elastic micro-vibration properties of three-dimensional physical space, providing a high-fidelity data source with strict physical interpretability for subsequent visual diagnosis.

[0045] Step S5: Equivalent superposition of environmental noise and automatic synthesis of video stream data.

[0046] On the original pixel displacement trajectory of the two-dimensional image plane obtained in step S4, multi-source environmental interference noise is equivalently superimposed. For example, Gaussian white noise is superimposed to simulate the thermal noise and shot noise of the virtual camera image sensor; simultaneously, a low-frequency large-amplitude drift signal is injected, the power spectral density of which follows a preset low-altitude gust disturbance model and a ground observation array sway interference model. By introducing the above environmental noise, a virtual pixel optical flow trajectory dataset containing local high-frequency elastic micro-vibration characteristics and macroscopic environmental interference is finally generated. A loop iteration mechanism is established in the parameter space to traverse different fault severity coefficients, multi-source physical damage fault combinations, and variable speed conditions, automatically batch synthesizing large-scale video stream datasets with strict physical interpretability and high fidelity.

[0047] Example 2 Building upon Example 1, this example provides another method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity. The difference lies in its application to health monitoring of turbine blade equipment. The underlying dynamic order reduction, parameterized fault injection, and geometric perspective projection mapping logic are completely consistent with Example 1. The physical entities and parameter connotations in each step are expanded accordingly as follows: In steps S1 and S2, the turbine rotor shaft and casing are simplified as a rigid matrix without elastic deformation. The cascaded blades and blade root tenon structure are treated as rotating working components and their flexible support / connection members. A three-dimensional finite element mesh model is established for the cascaded blades, and the orthogonal anisotropic or isotropic physical parameters of the material under high-temperature conditions are input. The Craig-Bampton modal synthesis method is used to extract the higher-order elastic characteristic modes of the blades, and modal neutral files are exported for multibody assembly. The contact and fit between the blade root tenon and the rotor disk are simulated using bushing force elements, and virtual displacement observation points are configured on the blade surface and blade tip edge.

[0048] In step S3, the boundary of the non-stationary operating condition is set as the time series curve of speed change during turbine start-up and shutdown or large load fluctuations, and the unsteady aerodynamic load is replaced with the high-temperature and high-pressure steam flow excitation force load based on complex flow channel calculations. By dynamically reducing the stiffness matrix of the bushing force element at the blade root tenon in the dynamic parameter space, physical damage faults such as blade root tenon loosening and blade crack degradation can be parameterized and simulated. Transient dynamic solutions are then run to extract the original three-dimensional spatial displacement time series data of each observation point on the blade.

[0049] In steps S4 and S5, the same virtual camera center perspective projection operator is used to map the three-dimensional spatial physical displacement vector of the blade and blade tip observation point into pixel displacement trajectories on the two-dimensional image plane, completely preserving the high-frequency blade tip vibration waveform. By equivalently superimposing sensor thermal noise on the pixel trajectory and injecting low-frequency large-amplitude drift signals caused by micro-amplitude shaking of the substrate and airflow disturbance, a high-fidelity video stream dataset for monitoring micro-vibration damage and fatigue state assessment of turbine blade tip clearance is automatically and batch-synthesized.

[0050] In summary, on the one hand, this invention, by introducing the Craig-Bampton modal synthesis method and bushing force elements to construct a rigid-flexible coupled benchmark dynamic model, overcomes the limitation of traditional multibody dynamics, which relies solely on rigid body modeling and cannot describe high-frequency elastic deformation. This allows for the accurate capture of local high-frequency micro-vibration characteristics of flexible support components and rotating working parts of rotating blade-type equipment under unsteady excitation. On the other hand, this invention constructs a virtual visual mapping operator based on three-dimensional perspective projection. During the mapping process, it completely preserves the high-frequency micro-vibration characteristics in the original displacement time-series data, ensuring that the physical feature signals caused by local high-frequency elastic micro-vibrations are not filtered out. This method establishes a quantitative cross-scale evolution model from microscopic physical damage faults to macroscopic two-dimensional image pixel displacements, completely filling the gap in existing visual diagnostic algorithms that lack physical damage mechanism support due to purely data-driven algorithms. Furthermore, without conducting high-cost, high-risk destructive experiments, this invention achieves comprehensive traversal and data synthesis of equipment across all operational boundary conditions and extreme conditions difficult to implement in experiments through parameterized dynamic injection of physical faults. This invention not only significantly reduces the cost of acquiring fault datasets and the complexity of physical experiments, but also provides rich samples for the training of subsequent visual diagnostic models, improving the generalization detection capability of visual diagnostic models for minor damage. Finally, in the automated synthesis of visual data, this invention not only superimposes Gaussian white noise to simulate the thermal noise of the camera sensor, but also simultaneously injects low-frequency, large-amplitude drift signals that conform to a preset low-frequency perturbation model to equivalently simulate real environmental interference such as low-altitude gusts and observation array swaying. The resulting video stream sequence, containing micro-vibration features and environmental interference, possesses extremely high physical fidelity and interpretability, effectively improving the generalization and anti-interference capabilities of visual diagnostic algorithms in harsh working conditions and low signal-to-noise ratio environments.

[0051] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity, characterized in that, Includes the following steps: S1, based on the rigid base model of the rotating blade equipment, performs rigid body geometry simplification and physical property equivalence to obtain the equivalent rigid base model and its physical property parameters, and performs free interface mode reduction for the flexible support / connection components in the vibration transmission path to extract multi-order regular modes. S2, based on the equivalent rigid body matrix model and the multi-order regular modes, multi-body topology assembly is performed to establish a nonlinear contact mechanics model between the supporting component and the supporting ground. In this model, a bushing force element is used to simulate the connection structure between the flexible component and the rigid matrix, and a stiffness matrix representing the initial connection state is assigned. The stiffness matrix is ​​dynamically adjusted in the parameter space to simulate the degradation process of the structural connection state. Observation markers are configured on the surface of the observable area of ​​the matrix to extract and output the dynamic response of the high-frequency micro-vibration of the structure, and a rigid-flexible coupling reference dynamic model of the whole machine is constructed. S3, set non-stationary working condition boundaries in the benchmark dynamic model to simulate the complex working conditions of the rotating blade equipment in actual operation, apply unsteady aerodynamic loads, inject physical damage faults into the system parameterization in the parameter space, solve the transient dynamics of operation, and calculate the original displacement time series data and acceleration time series data of the observation markers in three-dimensional space. S4, construct the virtual camera intrinsic parameter matrix, and use the principle of three-dimensional coordinate perspective projection to establish a mapping operator from three-dimensional physical space to two-dimensional image plane, and project the original displacement time series data into the original pixel displacement trajectory on the two-dimensional image plane. S5, environmental interference noise is superimposed on the original pixel displacement trajectory, and a low-frequency drift signal is injected to simulate environmental gust disturbance and observation array sway interference, generating a virtual pixel trajectory dataset containing micro-vibration characteristics and environmental interference. By iteratively traversing different fault severity coefficients, multi-source physical damage fault combinations and non-stationary working conditions in the parameter space, a video stream dataset is synthesized to establish a cross-scale evolution mapping relationship from macroscopic physical damage of equipment to microscopic pixel micro-vibration trajectory.

2. The method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity according to claim 1, characterized in that, In step S1, the process of obtaining the equivalent rigid body matrix model and physical property parameters includes the following steps: Remove internal non-load-bearing components and convert the total mass of electronic components into central counterweight parameters. Recalibrate and balance the total mass and center of gravity to obtain an equivalent rigid body matrix model and its physical property parameters. The process of extracting multi-order regular modes includes the following steps: For the flexible support / connection components in the vibration transmission path, a finite element mesh model is established, the orthogonal anisotropic physical parameters of the composite material are input, and the Craig-Bampton modal synthesis method is used to perform free interface mode reduction on the flexible component, and the first 15 to 20 canonical modes are extracted as the multi-order canonical modes.

3. The method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity according to claim 1, characterized in that, In S2, the nonlinear contact mechanics model is characterized by a bushing force element and a nonlinear contact model; wherein, the bushing force element includes a bushing force element used to simulate the connection state between the rotating blade and the motor shaft or transmission shaft, the bushing force element is used to simulate the bolt connection structure between the flexible component and the rigid substrate, the supporting component is the landing gear, and the nonlinear contact mechanics model is configured with a nonlinear contact stiffness coefficient, a damping coefficient and a Coulomb friction coefficient.

4. The method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity according to claim 1, characterized in that, In step S3, the process of injecting physical damage faults into the system parameterization within the parameter space includes the following steps: Lowering the diagonal coefficient value in the stiffness matrix of the bushing force element simulates the decrease in bolt preload and the degradation state of the connection. An eccentric mass block is added at a specified blade node of the rotating working part to generate a centrifugal excitation force proportional to the square of the rotational speed, which characterizes the asymmetric mass eccentricity.

5. The method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity according to claim 4, characterized in that, In S3, the non-stationary operating condition boundary is a preset rotating blade sweep frequency acceleration speed curve, and the unsteady aerodynamic load is a dynamic load calculated based on a quadratic aerodynamic model that changes nonlinearly with the rotational speed.

6. The method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity according to claim 1, characterized in that, In S4, the virtual camera intrinsic parameter matrix includes focal length, principal point coordinates, and distortion coefficients, and the three-dimensional coordinate perspective projection principle is the central perspective projection principle.

7. The method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity according to claim 1, characterized in that, In step S5, the environmental interference noise is Gaussian white noise, used to simulate the thermal noise of the camera sensor. The power spectral density of the low-frequency drift signal follows a preset low-frequency disturbance model, which includes a low-altitude gust disturbance model and a ground observation array sway interference model.

8. The method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity according to claim 1, characterized in that, The rotating blade equipment includes electric vertical takeoff and landing aircraft, steam turbines, wind turbines, aircraft engines, or industrial fans.

9. The method for synthesizing rotating blade video data based on rigid-flexible coupling visual mapping and physical fidelity according to claim 1, characterized in that, In step S5, the iterative traversal includes the following steps: By changing the fault severity coefficient within the parameter space and combining different physical damage faults, the system traverses various operating conditions, including non-stationary speeds, and synthesizes a video stream dataset.

10. An electronic device, characterized in that, It includes one or more processors, a memory, and one or more programs stored in the memory, said one or more programs including instructions for executing the rotating blade video data synthesis method based on rigid-flexible coupling visual mapping and physical fidelity as described in any one of claims 1-9.

Images