Drone structure flaw detection method and device based on voiceprint and spectrum, medium and product

Abstract

The application discloses a kind of based on voiceprint and spectrum unmanned plane structure flaw detection method, equipment, medium and product, it is related to structure flaw detection field.The method comprises the following steps: controlling unmanned plane flight and hovering at the target detection position of the structure to be measured;Controllable excitation device of unmanned plane is controlled to output mechanical excitation pulse to the structure to be measured;The original acoustic signal containing forced radiation sound wave and unmanned platform noise is collected by acoustic sensor array;The current real-time rotor speed data of unmanned plane is obtained, and it is used as dynamic environmental noise reference benchmark, and the original acoustic signal is processed to obtain target pure acoustic signal;Acoustic spectral feature extraction is carried out to target pure acoustic signal, and target acoustic characteristic parameter set is obtained;The target acoustic characteristic parameter set is input into the pre-trained structure damage identification model, and the damage determination result for the structure to be measured is obtained.The application can alleviate the problem of unmanned plane rotor noise interference acoustic detection.

Description

Technical Field

[0001] This invention relates to the field of structural flaw detection, and in particular to a method, equipment, medium, and product for structural flaw detection of unmanned aerial vehicles (UAVs) based on acoustic signatures and spectrum. Background Technology

[0002] Structural flaw detection refers to the process of inspecting and assessing the internal condition of building components such as wooden structures of houses and steel structures of tower cranes using appropriate testing methods to identify hidden damage such as cracks, voids, and insect infestation. With the continuous advancement of urban high-rise building, special equipment, and ancient building protection efforts, the demand for long-distance flaw detection of structures located at high altitudes, in dangerous areas, or inaccessible to personnel is increasing. Due to their flexible flight capabilities and good spatial accessibility, drones are gradually being introduced into this type of structural flaw detection operation.

[0003] In existing technologies, UAV structural flaw detection typically uses an airborne visible light camera or infrared thermal imager to acquire images of the structure's surface. Ground personnel then assess the structure's condition by observing visible features such as cracks, spalling, and color differences in the images. In addition, a few solutions attempt to equip UAVs with sound acquisition devices to passively receive sound wave signals generated by the structure under external environmental excitation and use the received sound wave signals to assist in judging the structure's condition.

[0004] However, during the acoustic flaw detection of structures using drones equipped with sound acquisition devices, the rotor noise generated by the drones during flight and hovering will simultaneously enter the sound acquisition device. This causes a large amount of noise components from the drone itself to be mixed in the collected sound wave signals, making it difficult to separate the effective sound wave signals that reflect the internal state of the structure and affecting the accurate identification of hidden damage inside the structure. Summary of the Invention

[0005] To address the aforementioned technical problems and deficiencies, the purpose of this invention is to provide a method, device, medium, and product for UAV structural flaw detection based on acoustic signature and spectrum, which can improve the accuracy of identifying hidden internal damage of structures using UAVs and alleviate the problem of UAV rotor noise interfering with acoustic flaw detection.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a method for structural flaw detection of unmanned aerial vehicles (UAVs) based on acoustic signature and spectrum, applied to a UAV control unit, the method comprising:

[0007] Control the drone to fly and hover at the target detection location of the structure to be tested;

[0008] The controllable excitation device of the UAV outputs mechanical excitation pulses to the structure under test to excite the structure under test to generate forced radiated sound waves.

[0009] The raw acoustic signals, including forced radiated sound waves and drone platform noise, are collected using the acoustic sensor array of the drone.

[0010] Obtain the current real-time rotor speed data of the drone;

[0011] Real-time rotor speed data is used as a reference for dynamic environmental noise, and noise cancellation processing is performed on the original acoustic signal to obtain the target pure acoustic signal.

[0012] Acoustic spectral features are extracted from the target clean acoustic signal to obtain the target acoustic feature parameter set;

[0013] The target acoustic feature parameter set is input into a pre-trained structural damage recognition model to obtain the damage determination result for the structure under test.

[0014] This invention employs the aforementioned method and steps, controlling a controllable excitation device mounted on a UAV to actively output mechanical excitation pulses to the structure under test. This excites the structure to generate forced radiated sound waves carrying internal structural state information, transforming the acoustic acquisition target from a passive sound wave under random environmental excitation to a controllable and repeatable active response sound wave. Simultaneously, the current real-time rotor speed data of the UAV is used as a dynamic environmental noise reference benchmark to perform noise cancellation processing on the original acoustic signal, ensuring that the noise suppression frequency band dynamically matches the rotor speed, thereby obtaining a target pure acoustic signal with a significantly improved signal-to-noise ratio. Based on this, acoustic spectral features are extracted from the target pure acoustic signal, and damage is determined using a pre-trained structural damage identification model. This achieves automated identification of hidden internal damage in structures at high altitudes or in areas difficult for personnel to access, improving the accuracy of identifying hidden internal damage and alleviating the problem of UAV rotor noise interfering with acoustic flaw detection.

[0015] In some implementations, acoustic spectral features are extracted from the target clean acoustic signal to obtain a set of target acoustic feature parameters, including:

[0016] Perform a Fast Fourier Transform on the target pure acoustic signal to obtain the frequency domain signal;

[0017] Extract the main resonant frequency parameter, high-frequency energy attenuation coefficient, and bandwidth energy ratio parameter from the frequency domain signal;

[0018] The target acoustic characteristic parameter set is obtained by splicing and combining the main resonant frequency parameter, the high-frequency energy attenuation coefficient, and the frequency band energy ratio parameter.

[0019] By employing the above technical solution, a fast Fourier transform is performed on the target pure acoustic signal, and three frequency domain feature parameters—the main resonant frequency parameter, the high-frequency energy attenuation coefficient, and the frequency band energy ratio parameter—are extracted from the frequency domain signal and then combined. The acoustic response characteristics are quantitatively described from three complementary dimensions: the position of the main resonant frequency, the high-frequency attenuation characteristics, and the frequency band energy distribution. This improves the acoustic features' ability to characterize and identify internal structural damage.

[0020] In some implementations, extracting the main resonant frequency parameter, the high-frequency energy attenuation coefficient, and the bandwidth energy ratio parameter from the frequency domain signal includes:

[0021] Extract the first integral total energy value of the frequency domain signal within the first preset frequency range;

[0022] Extract the second integral energy total value of the frequency domain signal within the second preset frequency interval, where the frequency band of the first preset frequency interval is higher than the frequency band of the second preset frequency interval.

[0023] The ratio of the first integral total energy value to the second integral total energy value is calculated to obtain the band energy ratio parameter.

[0024] By adopting the above technical solution, the total integral energy of the frequency domain signal in the first preset frequency interval of the higher frequency band and the second preset frequency interval of the lower frequency band are calculated respectively, and the ratio of the two is used as the frequency band energy ratio parameter. This gives the characteristic parameter a self-normalization characteristic for the fluctuation of excitation force and the difference of acquisition distance, thereby reducing the impact of excitation consistency deviation and changes in acquisition conditions on the stability of the characteristic parameter.

[0025] In some implementations, the controllable excitation device controlling the UAV outputs mechanical excitation pulses to the structure under test, including:

[0026] Control the drone to approach the structure under test until the flexible buffer mechanism at the front end of the controllable excitation device touches the surface of the structure under test;

[0027] The system acquires real-time compression data of the flexible buffer mechanism and, in response to the real-time compression data reaching a preset buffer threshold, controls the UAV to output anti-recoil compensation thrust using its own flight control system to maintain the contact state of the flexible buffer mechanism and the current hovering attitude.

[0028] While maintaining the current hovering attitude, the electromagnetic catapult component in the controllable excitation device outputs mechanical excitation pulses to the structure under test through a flexible buffer mechanism.

[0029] By adopting the above technical solution, the flexible buffer mechanism is controlled to contact the surface of the receiving structure, and the flight control system outputs anti-recoil compensation thrust after the real-time compression data reaches the preset buffer threshold. This enables the electromagnetic catapult assembly to output mechanical excitation pulses in a stable pre-compression state and hovering attitude, which improves the reliability of excitation force transmission and avoids attitude instability or separation of the UAV due to excitation reaction force.

[0030] In some embodiments, the electromagnetic ejection assembly within the controllable excitation device outputs mechanical excitation pulses to the structure under test via a flexible buffer mechanism, including:

[0031] Obtain the target material type of the structure to be tested;

[0032] If the target material type is the first material, the electromagnetic catapult assembly is controlled to output the first mechanical excitation pulse to the structure under test with the first driving current and the first pulse width;

[0033] If the target material type is the second material, the electromagnetic catapult assembly is controlled to output a second mechanical excitation pulse to the structure under test with a second driving current and a second pulse width.

[0034] Among them, the acoustic impedance of the first material is greater than that of the second material, the first driving current is greater than that of the second driving current, and the first pulse width is less than that of the second pulse width.

[0035] By adopting the above technical solution, the driving current and pulse width of the electromagnetic catapult component are adaptively matched according to the target material type of the structure under test. For the first material with high acoustic impedance, a combination of high current and short pulse width is used, and for the second material with low acoustic impedance, a combination of low current and long pulse width is used. This makes the mechanical excitation pulse compatible with the material characteristics of the structure under test, thus expanding the applicability of this method to building components of different material types.

[0036] In some implementations, the target acoustic feature parameter set is input into a pre-trained structural damage recognition model to obtain damage determination results for the structure under test, including:

[0037] The target acoustic feature parameter set is constructed into a one-dimensional feature vector and input into the structural damage identification model for feature mapping analysis;

[0038] Obtain the anomaly confidence probability values ​​for each candidate damage category output by the structural damage identification model;

[0039] If the maximum anomaly confidence probability value is greater than the preset alarm threshold, a damage determination result containing the internal damage type and the estimated damage size will be generated.

[0040] By adopting the above technical solution, the target acoustic feature parameter set is constructed into a one-dimensional feature vector and input into a classification neural network model for feature mapping analysis. Based on the comparison results of the abnormal confidence probability value of each candidate damage category with the preset alarm threshold, a damage judgment result containing internal damage type and estimated damage size is generated, realizing fine-grained automated judgment of damage type and damage degree.

[0041] In some implementations, the real-time rotor speed data includes the independent real-time speed data of each rotor unit when the UAV is currently in differential hovering state;

[0042] Real-time rotor speed data is used as a dynamic environmental noise reference, and noise cancellation processing is performed on the original acoustic signal to obtain the target clean acoustic signal, including:

[0043] Based on the real-time rotational speed data of each rotor unit and the number of blades configured in each rotor unit, the dynamic blade passing frequency and its harmonic frequencies of each rotor unit are calculated.

[0044] The dynamic blades corresponding to each rotor unit are aggregated by frequency and its harmonic frequencies to generate a multi-band dynamic noise reference matrix that is strongly correlated with the current hovering attitude.

[0045] The multi-band dynamic noise reference matrix is ​​used as a reference signal and input to the multi-reference source adaptive filter to track and cancel the aliased wandering aerodynamic noise generated by the differential operation of multiple independent rotors in the original acoustic signal, so as to obtain the target pure acoustic signal.

[0046] By adopting the above technical solution, the independent real-time rotational speed data of each rotor unit under differential hovering state are obtained separately. The dynamic blade passage frequency and harmonic frequency corresponding to each rotor unit are calculated and aggregated into a multi-band dynamic noise reference matrix. The matrix is ​​then input into a multi-reference source adaptive filter to track and cancel the aliased wandering aerodynamic noise generated by the differential operation of multiple independent rotors, thereby improving the cancellation accuracy and tracking capability of complex dynamic noise under differential hovering state.

[0047] In a second aspect, the present invention provides an electronic device comprising: one or more processors and a memory; the memory being coupled to the one or more processors, the memory being used to store computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the electronic device to perform the methods described in the first aspect and any possible implementation thereof.

[0048] Thirdly, the present invention provides a computer-readable storage medium storing computer instructions that, when executed on an electronic device, cause the electronic device to perform the methods described in the first aspect and any possible implementation thereof.

[0049] Fourthly, the present invention provides a computer program product including computer instructions that, when executed on an electronic device, cause the electronic device to perform the method described in the first aspect and any possible implementation thereof.

[0050] Understandably, the electronic device provided in the second aspect, the storage medium provided in the third aspect, and the computer program product provided in the fourth aspect are all used to execute the method provided by this invention. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here. Attached Figure Description

[0051] Figure 1 This is a schematic diagram illustrating an application scenario of a UAV structural flaw detection method based on acoustic signature and spectrum according to an embodiment of the present invention.

[0052] Figure 2 This is a flowchart illustrating a method for structural flaw detection of unmanned aerial vehicles based on acoustic signature and spectrum according to an embodiment of the present invention.

[0053] Figure 3 This is a schematic diagram of the electronic device hardware architecture of a drone control unit according to an embodiment of the present invention. Detailed Implementation

[0054] The terminology used in the following embodiments of the present invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the specification and appended claims of the present invention, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in the present invention refers to any or all possible combinations comprising one or more of the listed items.

[0055] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of the present invention, unless otherwise stated, "a plurality of" means two or more.

[0056] This invention provides a method for structural flaw detection in unmanned aerial vehicles (UAVs) based on acoustic signatures and spectrum analysis. This method is applied to the UAV control unit, which is a computing unit mounted on or communicatively connected to the UAV, used for data processing and control command generation.

[0057] like Figure 1 As shown, the control unit (hereinafter referred to as the "control unit") can be mounted on the main frame of the drone. Multiple rotors are respectively mounted at the ends of the various extension arms of the main frame of the drone. The high-speed rotation of the rotors provides the lift required for the drone to fly and hover. The aerodynamic noise generated by the rotors cutting through the air during rotation is the main source of noise for the drone platform. Figure 1 An arc-shaped arrow is marked above the rotor to indicate that the rotor is in a rotating working state.

[0058] The acoustic sensor array is connected to the main frame of the drone. The acoustic sensor array consists of multiple independent acoustic sensors arranged in combination.

[0059] The controllable vibration device is also connected to the main frame of the UAV. The front-end output of the controllable vibration device is positioned directly in front of the target detection location on the surface of the structure under test. During actual flaw detection operations, the control unit controls the UAV to fly and hover stably directly in front of the target detection location on the structure under test.

[0060] The control unit sends a trigger control command to the controllable vibration device. Upon receiving the trigger control command, the controllable vibration device outputs mechanical excitation pulses in a directional manner to the target detection position of the structure under test. Figure 1 The straight line segment with arrows is used to indicate the direction of application and the path of action of the mechanical excitation pulse.

[0061] When a mechanical excitation pulse is applied to the target detection position of the structure under test, it induces forced vibration in the structure at the target detection position. The forced vibration of the structure under test radiates outward into the external air medium through the surface of the structure under test, thus forming forced radiated sound waves that propagate outward.

[0062] Figure 1 Multiple concentric arcs are used to represent the physical process of forced radiated sound waves propagating from the surface of the structure under test into the surrounding space. An acoustic sensor array is positioned along the propagation path of the forced radiated sound waves. The acoustic sensor array is used to synchronously acquire sound wave signals in space. At this time, the sound wave signals in space not only include the forced radiated sound waves transmitted back from the surface of the structure under test, but also the noise from the UAV platform generated by the operation of multiple rotors above the UAV.

[0063] The acoustic sensor array transmits the raw acoustic signal, which is a mixture of drone platform noise and forced radiated sound waves, to the control unit.

[0064] The control unit acquires the real-time rotor speed data of each rotor and performs noise cancellation processing on the original acoustic signal based on the real-time rotor speed data to filter out the interference caused by rotor operation. Then, it extracts the target acoustic feature parameter set that can reflect the internal damage state of the structure under test and finally obtains the damage judgment result for the structure under test.

[0065] The following is combined with Figure 2 This embodiment provides a detailed description of the UAV structural flaw detection method based on acoustic signature and spectrum, specifically including the following steps:

[0066] Step S101: Control the drone to fly and hover at the target detection location of the structure to be tested.

[0067] The structure to be tested can specifically refer to building components that require internal damage inspection, including but not limited to exterior wall panels of high-rise buildings, steel structural components of tower cranes, structural components of bridges, and wooden structural components of ancient buildings, which are located at high altitudes or in areas difficult for personnel to directly access. The target detection location refers to a specific area on the surface of the structure to be tested that requires acoustic flaw detection, either pre-planned or specified in real-time by the operator.

[0068] In this step, the control unit, based on pre-set flight mission parameters or remote control commands sent by ground operators, controls the UAV to depart from the takeoff point and fly along the planned route to the vicinity of the spatial area where the structure under test is located. After the UAV arrives near the structure under test, the control unit further controls the UAV to adjust its flight attitude and spatial position, so that the UAV gradually approaches and eventually hovers at the target detection position of the structure under test.

[0069] Hovering refers to the flight state in which a drone maintains a relatively stable position and attitude in three-dimensional space. The control unit achieves flight control and hovering control through the drone's onboard flight control system. The flight control system comprehensively utilizes attitude data acquired by various airborne sensors, such as the inertial measurement unit, positioning module, barometric altimeter, and visual positioning sensor, to calculate the drone's current position and attitude in real time. It then continuously adjusts the rotational speed output of each rotor motor through a closed-loop control algorithm to maintain stable hovering of the drone at the target detection location.

[0070] After the drone hovers stably at the target detection location, it provides a stable aerial working platform for subsequent mechanical vibration operations and acoustic signal acquisition operations.

[0071] Step S102: Control the controllable excitation device of the UAV to output mechanical excitation pulses to the structure under test, so as to excite the structure under test to generate forced radiated sound waves.

[0072] Among them, the controllable vibration excitation device refers to a functional device mounted on the body of an unmanned aerial vehicle (UAV) that can be controlled by the control unit to apply a controllable transient mechanical impact force to the surface of the structure under test at the target detection location. The mechanical vibration pulse refers to the transient mechanical impact force output by the controllable vibration excitation device under the control of the control unit, which has a preset force amplitude and a preset duration. The mechanical vibration pulse acts on the surface of the structure under test in a contact or near-contact manner, applying a concentrated release of mechanical energy input to the structure under test within a short period of time.

[0073] Forced radiated sound waves refer to sound waves generated when a structure under test is subjected to a mechanical excitation pulse, resulting in forced vibrations due to the elastic properties of the structure's material. These forced vibrations radiate from the surface of the structure into the surrounding air, forming sound waves. Forced radiated sound waves carry information about the internal structural state of the structure under test. This is because the structural integrity of the structure directly affects its stiffness distribution and vibration modal characteristics, thus determining the frequency components and energy distribution characteristics of the forced radiated sound waves.

[0074] When the structure under test is free from damage such as cracks, voids, or erosion, the forced radiated sound waves exhibit acoustic spectral characteristics corresponding to the normal, intact state of the structure. When damage such as cracks, voids, or erosion exists within the structure under test, the presence of these defects alters the local stiffness distribution and vibration modes near the damaged area, causing a detectable shift or abnormal change in the acoustic spectral characteristics of the forced radiated sound waves relative to the normal, intact state.

[0075] In this step, after confirming that the UAV has stably hovered at the target detection position, the control unit sends an excitation execution command to the controllable excitation device. Upon receiving the excitation execution command, the controllable excitation device applies mechanical excitation pulses to the surface of the structure under test according to preset excitation parameters.

[0076] When a mechanical excitation pulse is applied to the surface of the structure under test, it generates elastic stress waves inside the structure. These elastic stress waves propagate in all directions within the structure and interact with its internal structure. During this process, the surface of the structure generates a vibration response related to the internal structural state. This vibration response drives pressure fluctuations in the air medium near the surface of the structure, thereby forming forced radiation sound waves that propagate into the surrounding space.

[0077] By actively applying mechanical excitation pulses to the structure under test, forced radiated sound waves can be generated at a controllable time and with a controllable excitation intensity. Compared with passive acquisition methods that rely on random excitation sources such as wind load and traffic vibration in the external environment, the active excitation method makes the generation time and excitation conditions of forced radiated sound waves controllable, thereby improving the repeatability and consistency of acoustic signal acquisition.

[0078] Step S103: Collect raw acoustic signals containing forced radiated sound waves and UAV platform noise through the acoustic sensor array of the UAV.

[0079] An acoustic sensor array refers to a sound acquisition device mounted on the airframe of a drone, consisting of multiple acoustic sensors arranged in a preset spatial geometry. Each acoustic sensor in the array converts sound wave signals propagating in the air into corresponding analog electrical signals, which are then processed by analog-to-digital conversion to form digital acoustic signals. Using an array of multiple acoustic sensors for sound acquisition, compared to using a single acoustic sensor, allows for enhanced signal reception from specific directions through array signal processing technology.

[0080] Unmanned aerial vehicle (UAV) platform noise refers to the sum of various noise components generated by the UAV during hovering, including aerodynamic noise from the UAV's rotor cutting through the air, mechanical noise from the rotor motors, and radiated noise from the vibration of the airframe structure. UAV platform noise persists continuously while the UAV is in flight or hovering, and its frequency and amplitude vary with rotor speed.

[0081] The raw acoustic signal refers to the acoustic signal collected by the acoustic sensor array without noise removal processing. The raw acoustic signal contains both forced radiation acoustic wave components and UAV platform noise components.

[0082] In this step, the control unit controls the acoustic sensor array to perform acoustic signal acquisition within a preset acquisition time window after the controllable excitation device outputs a mechanical excitation pulse. The preset acquisition time window refers to the time interval from the moment the mechanical excitation pulse is applied until the forced radiated sound wave attenuates to an undetectable level.

[0083] Each acoustic sensor in the acoustic sensor array synchronously receives sound wave signals propagating in the surrounding air medium within a preset acquisition time window. Since the rotor of the UAV continues to rotate to provide lift during hovering, UAV platform noise inevitably exists in the acoustic acquisition environment of the acoustic sensor array. Therefore, the raw acoustic signal acquired by the acoustic sensor array contains both forced radiated sound wave components reflecting the internal structural state information of the structure under test and UAV platform noise components originating from the operation of the UAV itself.

[0084] The acoustic sensor array collects raw acoustic signals and transmits them to the control unit in the form of digital signals after analog-to-digital conversion, for processing in subsequent steps.

[0085] Step S104: Obtain the current real-time rotor speed data of the UAV.

[0086] The real-time rotor speed data includes the current real-time rotational speed of each rotor of the UAV within the same time period during which the acoustic sensor array collects the raw acoustic signals. The rotor speed directly determines the frequency and amplitude characteristics of the aerodynamic noise components in the UAV platform noise. The higher the rotor speed, the higher the fundamental frequency and harmonic frequencies of the aerodynamic noise generated by the rotor rotation, and the greater the noise amplitude. The lower the rotor speed, the lower the corresponding noise frequency and the smaller the noise amplitude.

[0087] In this step, the control unit reads the real-time rotational speed data of each rotor from the drone's flight control system. The flight control system monitors the current rotational speed of each rotor motor in real time through the speed feedback interface built into the electronic speed controller or independent rotor speed sensors, and sends the real-time rotational speed data of each rotor to the control unit.

[0088] In order to maintain the stability of spatial position and flight attitude when the drone is hovering, the flight control system needs to continuously and dynamically adjust the speed output of each rotor motor according to factors such as external wind disturbances and load changes caused by the movement of airborne equipment. Therefore, the speed of each rotor is in a state of real-time change.

[0089] The purpose of acquiring real-time rotor speed data is to provide reference information directly related to the current noise characteristics of the UAV platform for subsequent noise cancellation processing. There is a deterministic physical correspondence between the frequency composition of the most significant aerodynamic noise component of the UAV platform and the rotor speed, which can be derived from aerodynamic principles. Therefore, by acquiring real-time rotor speed data, the frequency characteristics of the UAV platform noise at the current moment can be dynamically calculated.

[0090] Step S105: Use the real-time rotor speed data as a reference for dynamic environmental noise, and perform noise cancellation processing on the original acoustic signal to obtain the target pure acoustic signal.

[0091] In this step, the description of step S105 will be optimized and improved, and the implementation details of this step will be explained in more detail.

[0092] Step S105: Use the real-time rotor speed data as a reference for dynamic environmental noise, and perform noise cancellation processing on the original acoustic signal to obtain the target pure acoustic signal.

[0093] Among them, the dynamic environmental noise reference benchmark refers to the reference signal or reference feature data calculated based on real-time rotor speed data, which is used to characterize the noise distribution characteristics of the UAV platform in the frequency domain at the current moment.

[0094] Since the rotational speed of each rotor of the UAV changes continuously during hovering due to the real-time control of the flight control system, the frequency structure and energy distribution of the corresponding UAV platform noise also exhibit real-time changing characteristics. Therefore, the reference information constructed based on the real-time rotor speed data also needs to be dynamically updated synchronously with the changes in the real-time rotor speed data, hence it is called the dynamic environmental noise reference benchmark.

[0095] Noise cancellation processing refers to the signal processing process that uses the noise characteristic information provided by the dynamic environmental noise reference standard to identify and cancel the signal components belonging to the noise of the UAV platform from the original acoustic signal.

[0096] The target pure acoustic signal refers to the acoustic signal obtained after noise cancellation processing of the original acoustic signal, in which the noise component of the UAV platform is significantly suppressed while the forced radiated sound wave component is preserved.

[0097] In this step, the control unit's processing procedure specifically includes the following sub-steps:

[0098] (1) Conversion of rotor speed to noise characteristic frequency.

[0099] The dominant component of noise in UAV platforms is aerodynamic noise generated by the rotor cutting through the air, and there is a definite physical correspondence between the frequency characteristics of aerodynamic noise and the rotor rotation frequency.

[0100] Specifically, when the rotor rotates at a speed of several revolutions per minute, the control unit first converts the real-time rotor speed data (in revolutions per minute) into a rotor rotation frequency (in Hertz). The conversion relationship is that the rotor rotation frequency equals the real-time rotor speed data divided by sixty. The aerodynamic noise generated by the rotor rotating in the air mainly manifests as discrete frequency components at the fundamental frequency and several harmonics related to the rotor rotation frequency, accompanied by broadband aerodynamic noise components covering a certain frequency band.

[0101] Based on the converted rotor rotation frequency, the control unit further calculates the specific frequency position of each discrete frequency component in the UAV platform noise. The calculated discrete frequency positions together constitute the frequency characteristic description of the UAV platform noise at the current moment.

[0102] (2) Construction of dynamic environmental noise reference standard.

[0103] The control unit constructs a dynamic environmental noise reference benchmark for subsequent noise cancellation processing based on the frequency characteristics of the UAV platform noise calculated in the first sub-stage.

[0104] Specifically, the control unit generates sinusoidal and cosine reference waveforms for each calculated discrete frequency position. These sinusoidal and cosine reference waveforms are then discretized and sampled at the same sampling rate as the original acoustic signal to obtain a time-domain reference signal sequence. The frequency position of the generated reference signal sequence in the frequency domain corresponds to the main frequency components of the UAV platform noise at the current moment, thus characterizing the frequency structure features of the UAV platform noise.

[0105] When the real-time rotor speed data changes, the control unit synchronously updates the calculated discrete frequency position and regenerates the reference signal sequence accordingly, thereby enabling the dynamic environmental noise reference to track the real-time rotor speed change.

[0106] The construction of the dynamic environmental noise reference benchmark is a continuous process, ensuring that the dynamic environmental noise reference benchmark always corresponds to the actual rotor speed at the current moment throughout the entire period of original acoustic signal acquisition.

[0107] (3) Implementation of noise cancellation treatment.

[0108] The control unit employs an adaptive filtering algorithm to cancel noise in the original acoustic signal. The adaptive filtering algorithm uses a dynamic environmental noise reference as the reference input signal and the original acoustic signal as the input signal to be processed. By adaptively adjusting the weight coefficients of the filter, the estimated noise signal output after the reference input signal is mapped by the filter approximates the actual UAV platform noise components contained in the original acoustic signal in terms of waveform.

[0109] Specifically, at each sampling time, the adaptive filtering algorithm multiplies the current dynamic environmental noise reference by the current filter weight coefficient to obtain the estimated noise signal at the current time. Then, it subtracts the estimated noise signal from the original acoustic signal to obtain the residual signal at the current time. The adaptive filtering algorithm takes minimizing the energy of the residual signal as its optimization objective and updates and adjusts the filter weight coefficients according to the recursive iteration rule, so that the estimated noise signal obtained at subsequent sampling times can more accurately match the UAV platform noise component in the original acoustic signal.

[0110] The adaptive filtering algorithms that can be used include any one of the following: least mean square algorithm, normalized least mean square algorithm, and recursive least square algorithm.

[0111] Since there is no correlation in frequency structure between the forced radiated sound wave and the dynamic environmental noise reference, the adaptive filtering algorithm cannot map the forced radiated sound wave outside the residual signal during the iterative adjustment of the filter weight coefficients. Therefore, the forced radiated sound wave component will remain in the residual signal.

[0112] The UAV platform noise component in the original acoustic signal is strongly correlated with the dynamic environmental noise reference standard. It will be identified by the adaptive filtering algorithm and subtracted from the original acoustic signal by estimating the noise signal, thereby being suppressed in the residual signal.

[0113] (4) Output and quality verification of the target pure acoustic signal.

[0114] The control unit saves the residual signal output by the adaptive filtering algorithm as the target pure acoustic signal. To ensure the quality of the target pure acoustic signal, the control unit can also evaluate the signal-to-noise ratio of the target pure acoustic signal. Specifically, it calculates the total energy of the target pure acoustic signal and the residual energy of the target pure acoustic signal at each discrete frequency position corresponding to the dynamic ambient noise reference, and uses the ratio of the two as the signal-to-noise ratio evaluation index.

[0115] When the signal-to-noise ratio evaluation index is lower than the preset signal-to-noise ratio qualified threshold, the control unit determines that the effect of the noise cancellation process has not met the requirements, and returns to step S102 to re-trigger mechanical excitation and collect acoustic signals.

[0116] When the signal-to-noise ratio (SNR) evaluation index is not lower than the preset SNR qualification threshold, the control unit determines that the effect of this noise cancellation process meets the requirements and sends the target clean acoustic signal to the subsequent steps for acoustic spectrum feature extraction.

[0117] Through the coordinated execution of the above sub-steps, the control unit dynamically constructs a dynamic environmental noise reference benchmark that matches the current frequency characteristics of the UAV platform noise using real-time rotor speed data. Based on this dynamic environmental noise reference benchmark, the original acoustic signal is processed by an adaptive filtering algorithm to cancel noise, thereby effectively suppressing the noise component caused by the UAV's own rotor rotation in the original acoustic signal, while retaining the forced radiated sound wave component that reflects the internal structural state of the structure under test. Finally, a target clean acoustic signal with a significantly improved signal-to-noise ratio is obtained, providing a high-quality signal foundation for subsequent acoustic spectrum feature extraction and structural damage identification.

[0118] Step S106: Extract acoustic spectral features from the target pure acoustic signal to obtain the target acoustic feature parameter set.

[0119] Acoustic spectral feature extraction refers to the frequency domain analysis and processing of a target pure acoustic signal to extract multiple quantitative feature parameters that characterize the acoustic response of the structure under test from its spectrum. The target acoustic feature parameter set is a set of parameters composed of multiple feature parameters obtained through acoustic spectral feature extraction. This set comprehensively describes the distribution characteristics of the target pure acoustic signal in the frequency domain in a numerical form.

[0120] In this step, the control unit first performs a time-domain to frequency-domain transformation on the target pure acoustic signal, converting it from a time-domain representation of amplitude variation over time to a frequency-domain representation of amplitude variation over frequency, thereby obtaining the spectrum of the target pure acoustic signal. The spectrum of the target pure acoustic signal reflects the amplitude and energy distribution of each frequency component in the target pure acoustic signal.

[0121] After obtaining the spectrum of the target pure acoustic signal, the control unit extracts multiple characteristic parameters from the spectrum that reflect the internal structural state of the structure under test. These characteristic parameters characterize the spectral properties of the forced radiated sound waves generated by the structure under test after being subjected to a mechanical excitation pulse, from different frequency domain dimensions.

[0122] When the internal structure of the structure under test is in a normal and intact state, the spectrum of the forced radiated sound wave exhibits a specific spectral distribution pattern corresponding to this normal and intact state, and the various feature parameters extracted from the spectrum are within the numerical range corresponding to the normal and intact state.

[0123] When the structure under test has damage defects such as cracks, voids, or erosion, these defects alter the local stiffness and vibration propagation characteristics of the structure in the damaged area, leading to changes in the spectral distribution pattern of the forced radiated sound waves. For example, the presence of cracks may cause an increase or decrease in the energy of certain frequency components, the presence of voids may alter the resonance characteristics of specific frequency bands, and the presence of erosion may cause a local decrease in material density, thus affecting the sound wave propagation characteristics.

[0124] These changes ultimately manifest as the feature parameters extracted from the spectrum deviating from the numerical range of the normal, intact state. The control unit aggregates the extracted feature parameters into a target acoustic feature parameter set according to a preset combination method. The target acoustic feature parameter set constitutes a complete numerical feature representation describing the acoustic response characteristics of the structure under test at the current target detection location, providing a standardized data format for the input of the structural damage identification model in subsequent steps.

[0125] Step S107: Input the target acoustic feature parameter set into the pre-trained structural damage recognition model to obtain the damage judgment result for the structure under test.

[0126] In this step, the structural damage identification model refers to a machine learning model that has been trained in advance using a large amount of training sample data with labeled damage states. The training sample data includes multiple sets of acoustic feature parameters from different structures under different damage states, as well as damage state labeling information corresponding to each set of acoustic feature parameters.

[0127] Damage state annotation information characterizes the actual internal damage status of the structure from which the corresponding training sample originates during acoustic signal acquisition. This information covers both undamaged states and different types of damage. Through learning from a large amount of training sample data, the structural damage identification model establishes a mapping relationship between acoustic feature parameters and the internal damage status of the structure. This enables the model, after training, to infer the internal damage status of the corresponding structure based on the input acoustic feature parameter set.

[0128] Damage assessment results refer to the conclusions of the structural damage identification model regarding the existence and extent of damage within the structure under test, based on the input set of target acoustic feature parameters.

[0129] In this step, the control unit sends the target acoustic feature parameter set obtained in step S106 as input data to the pre-trained structural damage identification model for inference calculation. After receiving the target acoustic feature parameter set, the structural damage identification model analyzes the values ​​of each feature parameter in the target acoustic feature parameter set and the combination relationships between each feature parameter based on the mapping relationship between acoustic feature parameters and structural damage state established during the training phase, and generates a damage judgment result for the structure under test accordingly.

[0130] The training sample data used in the training phase of the structural damage identification model covers acoustic feature parameter samples of various structural materials, various damage types, and no-damage states, enabling the structural damage identification model to distinguish and identify various different structural damage states.

[0131] The damage assessment result at least indicates whether there is internal damage in the structure under test at the target detection location. When it is determined that there is internal damage in the structure under test, the damage assessment result can further include descriptive information about the damage.

[0132] After obtaining the damage assessment results, the control unit can store the damage assessment results in the local storage medium, or send the damage assessment results in real time to the monitoring terminal equipment on the ground for display and output through the wireless communication link between the UAV and the ground end. This allows on-site personnel to know the flaw detection results of the structure under test at the current target detection location in a timely manner, and decide whether further manual review or repair is required based on the damage assessment results.

[0133] This embodiment employs the above-described method and steps, controlling a UAV equipped with a controllable excitation device to actively apply mechanical excitation pulses to the structure under test, thereby controlling the structure under test to generate forced radiated sound waves carrying internal structural state information, and using an airborne acoustic sensor array to collect the forced radiated sound waves. Compared with passive acoustic acquisition methods that rely on external random environmental excitation, the excitation time and excitation conditions of this method are controllable and repeatable, improving the stability and consistency of acoustic flaw detection.

[0134] Meanwhile, this embodiment uses the real-time rotor speed data of the UAV to construct a dynamic environmental noise reference benchmark, and performs targeted noise cancellation processing on the UAV platform noise in the original acoustic signal accordingly. This effectively solves the technical problem that rotor aerodynamic noise seriously interferes with the effective acoustic signal when the UAV is equipped with an acoustic acquisition device to perform aerial flaw detection operations. This allows the target pure acoustic signal separated from the noise cancellation processing to more accurately reflect the acoustic response characteristics of the structure under test.

[0135] Building upon this foundation, this embodiment further extracts acoustic spectral features from the target pure acoustic signal and inputs them into a pre-trained structural damage identification model for intelligent identification. This achieves automated judgment of hidden damage inside the structure under test, reduces reliance on the professional experience of operators, and improves the accessibility and detection efficiency of structural flaw detection for building components in high-altitude or inaccessible areas.

[0136] In this embodiment, the process of building and training the structural damage identification model can be referred to as follows.

[0137] 1. The process of building a structural damage identification model includes determining the network structure of the model and configuring the training parameters.

[0138] In terms of network structure, the structural damage recognition model adopts a multi-layer feedforward neural network architecture, which includes an input layer, multiple hidden layers, and an output layer.

[0139] The number of nodes in the input layer corresponds to the number of feature parameters contained in the target acoustic feature parameter set, and is used to receive the target acoustic feature parameter set represented in numerical form.

[0140] Each of the multiple hidden layers contains a preset number of neurons. The layers are connected by a fully connected method. Each neuron in each layer transforms the weighted input value through a non-linear activation function to achieve a layer-by-layer abstract representation and non-linear mapping of the input features.

[0141] The number of nodes in the output layer corresponds to the number of preset candidate damage categories. Each output node corresponds to one candidate damage category, including no damage and various specific damage type categories, such as crack damage, void damage, and erosion damage. The output layer uses a normalized exponential function to convert the values ​​of each output node into a probability distribution. The output value of each output node represents the probability that the input sample belongs to the corresponding candidate damage category.

[0142] 2. The training process of the structural damage identification model includes two stages: the construction of the training dataset and the iterative optimization of the model parameters.

[0143] (1) The construction phase of the training dataset:

[0144] Training sample data was obtained by conducting acoustic excitation and response acquisition experiments on structural specimens of various material types. The structural specimens included undamaged specimens in a normal and intact state as well as artificially prefabricated specimens with damage defects of different types and sizes.

[0145] For each structural specimen, the same process as steps S102 to S106 of this embodiment can be used to perform mechanical excitation, acoustic signal acquisition, noise cancellation processing, and acoustic spectrum feature extraction operations, thereby obtaining the acoustic feature parameter set corresponding to the structural specimen in the current damage state.

[0146] At the same time, based on the actual damage condition of the structural specimen, the acoustic feature parameter set is labeled with the corresponding damage status label. The damage status label indicates whether the specimen belongs to the no-damage category or a specific damage type category.

[0147] The acoustic feature parameter sets corresponding to each structural specimen are paired with damage state labels to form training samples, and multiple training samples are aggregated to form a training dataset. The training dataset includes a large number of training samples collected under different material types, different damage types, and different damage sizes to ensure that the training dataset has sufficient diversity and representativeness.

[0148] (2) Iterative optimization stage of model parameters:

[0149] The training samples in the training dataset are input into the structural damage identification model in batches for forward propagation calculation to obtain the model's prediction output for each training sample.

[0150] The model's predicted output is compared with the true damage state labels of the training samples, and the cross-entropy loss value is calculated. The cross-entropy loss value quantifies the degree of deviation between the model's current predicted output and the true label. Based on the calculated cross-entropy loss value, the backpropagation algorithm is used to calculate the gradient of the loss value with respect to the connection weight parameters of each layer in the model, and the gradient descent optimization algorithm is used to update the connection weight parameters of each layer according to the gradient direction, so that the cross-entropy loss value gradually decreases.

[0151] The aforementioned forward propagation, loss calculation, backpropagation, and parameter update processes are repeatedly iterated on the training dataset until the cross-entropy loss value converges to below the preset convergence threshold or the number of iterations reaches the preset maximum number of iterations, at which point the training process ends.

[0152] After training, the connection weight parameters of each layer of the structural damage recognition model are fixed and saved to form a deployable pre-trained model file. This pre-trained model file is loaded into the memory of the UAV control unit for inference calculation in step S107 of this method.

[0153] This invention also provides a method for structural flaw detection of unmanned aerial vehicles based on acoustic signature and spectrum, comprising the following steps:

[0154] Step S201: Control the drone to fly and hover at the target detection location of the structure to be tested.

[0155] The specific implementation method of this step can be referred to step S101 of the aforementioned embodiment, and will not be repeated here.

[0156] Step S202: Control the drone to approach the structure under test until the flexible buffer mechanism at the front end of the controllable excitation device touches the surface of the structure under test.

[0157] The flexible buffer mechanism refers to a contact buffer component made of a material with elastic deformation capability, installed at the front end of the controllable vibration device. When not subjected to external force, the flexible buffer mechanism remains in a naturally extended state. When the front end of the flexible buffer mechanism is compressed by external force, it can undergo elastic compressive deformation along the direction of the force, and can return to its naturally extended state after the external force is removed. The function of the flexible buffer mechanism is to absorb the impact force generated at the moment of contact when the UAV approaches the structure under test until physical contact occurs, preventing a rigid collision between the UAV body and the structure under test that could cause a sudden disturbance to the UAV's attitude stability, and also preventing unexpected mechanical damage to the surface of the structure under test.

[0158] In this step, after confirming that the drone has hovered stably at the target detection position, the control unit further controls the drone to slowly translate along the direction toward the surface of the structure to be tested, so that the drone gradually reduces the distance between itself and the surface of the structure to be tested.

[0159] As the drone approaches the structure under test, the flexible buffer mechanism installed at the front end of the controllable vibration device is the first component to come into contact with the structure, making physical contact with its surface. The drone continues to advance towards the structure until the front end of the flexible buffer mechanism stably abuts against its surface.

[0160] The contact state refers to the continuous physical contact between the front end of the flexible buffer mechanism and the surface of the structure under test, and the flexible buffer mechanism undergoes a certain degree of elastic compression deformation under the action of contact pressure.

[0161] Step S203: Obtain real-time compression data of the flexible buffer mechanism, and in response to the real-time compression data reaching the preset buffer threshold, control the UAV to output anti-recoil compensation thrust using its own flight control system to maintain the contact state of the flexible buffer mechanism and the current hovering attitude.

[0162] Real-time compression data refers to the real-time measurement of the elastic compression deformation of the flexible buffer mechanism relative to its natural extension state along the direction of force at the current moment. This real-time compression data is acquired through displacement sensors installed inside or associated with the flexible buffer mechanism. The preset buffer threshold is a pre-set limit value used to determine if the flexible buffer mechanism has generated sufficient compression.

[0163] When the real-time compression data reaches the preset buffer threshold, it indicates that the flexible buffer mechanism has been compressed to the extent that an effective force transmission path is formed between the excitation output end of the controllable excitation device and the surface of the structure under test. In other words, the flexible buffer mechanism has sufficient pre-compression force to ensure that subsequent mechanical excitation pulses can be reliably transmitted to the surface of the structure under test through the flexible buffer mechanism.

[0164] Anti-recoil compensation thrust refers to the additional thrust component generated by the UAV flight control system after detecting that the flexible buffer mechanism has formed contact with the surface of the structure under test, and which is directed towards the surface of the structure under test. The purpose of anti-recoil compensation thrust is to counteract the reverse thrust effect on the UAV body generated by the reaction force generated when the electromagnetic catapult component outputs mechanical excitation pulses in subsequent steps, preventing the UAV from being ejected from the surface of the structure under test or from becoming unstable under the action of the reaction force of the mechanical excitation pulses.

[0165] In this step, the control unit continuously reads the real-time compression data of the flexible buffer mechanism fed back by the displacement sensor and compares the real-time compression data with the preset buffer threshold.

[0166] When the control unit determines that the real-time compression data has reached a preset buffer threshold, it sends an anti-recoil compensation command to the flight control system. Upon receiving the anti-recoil compensation command, the flight control system adds an additional anti-recoil compensation thrust output towards the surface of the structure under test, on top of the basic control loop that maintains the current hovering position and attitude. The magnitude of the anti-recoil compensation thrust is pre-calibrated based on the maximum expected reaction force generated when the controllable excitation device outputs a mechanical excitation pulse.

[0167] By outputting anti-recoil compensation thrust through the flight control system, the UAV can maintain the contact between the flexible buffer mechanism and the surface of the structure under test when performing mechanical excitation pulse output operations, while maintaining the current hovering attitude of the UAV without significant deviation.

[0168] In this embodiment, the current hovering attitude refers to the spatial position and flight attitude of the UAV when the flexible buffer mechanism abuts against the surface of the structure to be tested and the real-time compression data reaches the preset buffer threshold.

[0169] Step S204: While maintaining the current hovering attitude, control the electromagnetic catapult component in the controllable excitation device to output mechanical excitation pulses to the structure under test through the flexible buffer mechanism.

[0170] The electromagnetic catapult assembly refers to an actuator installed inside a controllable vibration device that converts electrical energy into transient mechanical impact force using the principle of electromagnetic force. The electromagnetic catapult assembly includes an electromagnetic coil and a catapult impact component that can be driven to move along the axial direction of the electromagnetic coil.

[0171] When a driving current is applied to the electromagnetic coil, a strong pulsed magnetic field is generated at the location of the ejector impactor. The strong pulsed magnetic field applies an electromagnetic driving force to the ejector impactor, pushing it to move at high speed along the axial direction and impact the inner end face of the flexible buffer mechanism.

[0172] The impact force exerted by the projectile impactor on the inner end face of the flexible buffer mechanism is transmitted to the surface of the structure under test through the flexible buffer mechanism, forming a mechanical excitation pulse acting on the surface of the structure under test. The flexible buffer mechanism plays a role in pulse shaping during the force transmission process, converting the rigid impact force of the projectile impactor into a force pulse waveform with a certain rise time and duration, thus avoiding excessively sharp force pulses that may cause localized damage to the surface of the structure under test.

[0173] In this step, after confirming that the flight control system has output anti-recoil compensation thrust and that the UAV maintains its current hovering attitude, the control unit sends an excitation trigger signal to the electromagnetic catapult component in the controllable excitation device.

[0174] After receiving the excitation trigger signal, the electromagnetic catapult assembly generates a pulsed electromagnetic field under the control of preset driving current and preset pulse width parameters, driving the catapult impact component to impact the inner end face of the flexible buffer mechanism at high speed.

[0175] The impact force is transmitted to the surface of the structure under test through a flexible buffer mechanism, forming a mechanical excitation pulse. The mechanical excitation pulse excites elastic stress waves inside the structure under test. After the elastic stress waves propagate inside the structure under test and interact with the material structure inside the structure, the surface of the structure under test generates a vibration response and radiates forced radiated sound waves into the surrounding air medium.

[0176] During the process of the electromagnetic catapult assembly outputting mechanical excitation pulses, while the catapult impact component applies an impact force to the flexible buffer mechanism, the reaction force generated by Newton's third law acts on the UAV body. The anti-recoil compensation thrust output by the flight control system is used to counteract the reverse thrust effect of this reaction force on the UAV body, so that the UAV maintains its current hovering attitude stability throughout the entire excitation process.

[0177] Step S205: Acquire raw acoustic signals using the acoustic sensor array of the UAV.

[0178] The specific implementation method of this step can be referred to step S103 of the aforementioned embodiment, and will not be repeated here.

[0179] Step S206: Obtain the current real-time rotor speed data of the UAV.

[0180] In this step, the real-time rotor speed data specifically includes the independent real-time speed data of each rotor unit when the UAV is currently in differential hovering state.

[0181] Differential hovering refers to the hovering state in which, during the hovering process of a UAV, the flight control system outputs different speed commands to each rotor unit to maintain flight attitude stability or to withstand the eccentric load from the surface of the receiving structure due to the flexible buffer mechanism, resulting in each rotor unit rotating at its own different speed simultaneously.

[0182] A rotor unit refers to the unit consisting of each independently operating rotor on a UAV and the motor that drives the rotor to rotate.

[0183] In differential hovering, the rotational speeds of the individual rotor units are inconsistent, resulting in differences in the frequency of the aerodynamic noise generated by each rotor unit. The aerodynamic noise generated by each rotor unit manifests in the frequency domain as multiple noise components with different frequency positions that change in real time.

[0184] In this step, the control unit reads the independent real-time rotational speed data of each rotor unit from the flight control system. Because the UAV is in a differential hovering state, there are differences in the real-time rotational speed data of each rotor unit, and the rotational speed of each rotor unit continuously changes with the real-time adjustments of the flight control system. The control unit records and saves the independent real-time rotational speed data of each rotor unit separately, so that subsequent steps can be used to calculate the corresponding noise frequency characteristics for each rotor unit.

[0185] Step S207: Based on the real-time rotational speed data of each rotor unit and the number of blades configured in each rotor unit, calculate the dynamic blade passing frequency and its harmonic frequencies for each rotor unit.

[0186] Among them, the blade number parameter refers to the number of blades configured on the rotor of each rotor unit. The blade number parameter is a fixed parameter determined at the factory of each rotor unit and is pre-stored in the memory of the control unit.

[0187] Dynamic blade passage frequency refers to the frequency value corresponding to the number of times a rotor unit's blades pass through a fixed observation point in space per unit time at the current real-time rotational speed. The dynamic blade passage frequency is calculated by multiplying the real-time rotational speed data of the rotor unit by the number of blades configured in that rotor unit. Because the rotational speed of the rotor unit changes in real time, the dynamic blade passage frequency also changes in real time, hence the name dynamic blade passage frequency.

[0188] Harmonic frequencies refer to integer multiples of the dynamic blade passage frequency, including the second, third, and higher integer multiples of the dynamic blade passage frequency. The energy of the aerodynamic noise generated by the rotor unit during rotation is mainly concentrated at the frequency positions corresponding to the dynamic blade passage frequency and its various harmonic frequencies.

[0189] In this step, the control unit multiplies the real-time rotational speed data of each rotor unit with the blade number parameter corresponding to each rotor unit to calculate the dynamic blade passage frequency of each rotor unit.

[0190] Based on this, the control unit further multiplies the dynamic blade passage frequency of each rotor unit by a preset integer harmonic order sequence to calculate the corresponding harmonic frequencies of each rotor unit.

[0191] Because each rotor unit operates at a differential speed, the real-time rotational speed data of each rotor unit is different. Therefore, the dynamic blade passing frequency and the frequencies of each harmonic order calculated by each rotor unit are different in value.

[0192] Step S208: Aggregate the dynamic blades corresponding to each rotor unit by frequency and its harmonic frequencies to generate a multi-band dynamic noise reference matrix that is strongly correlated with the current hovering attitude.

[0193] Among them, the multi-band dynamic noise reference matrix refers to a matrix-form reference data structure containing multiple frequency channels, which is formed by converging and integrating the dynamic blades of each rotor unit through the frequency and the harmonic frequencies of each order.

[0194] Each row in the multi-band dynamic noise reference matrix corresponds to a rotor unit, and each row contains the values ​​of the dynamic blade passage frequency and the harmonic frequencies of each order of the rotor unit. The multi-band dynamic noise reference matrix comprehensively describes the multi-band distribution characteristics of the UAV platform noise generated by all rotor units in the frequency domain under the current differential hovering state.

[0195] Since the rotational speed of each rotor unit changes continuously as the flight control system maintains the current hovering attitude in real time, the values ​​of each frequency channel in the multi-band dynamic noise reference matrix are also updated in real time. Therefore, there is a strong correlation between the multi-band dynamic noise reference matrix and the current hovering attitude.

[0196] In this step, the control unit arranges and aggregates the dynamic blade pass frequencies and harmonic frequencies of each rotor unit calculated in the previous step according to the rotor unit number and harmonic order, constructing a multi-band dynamic noise reference matrix. The number of rows in the multi-band dynamic noise reference matrix is ​​equal to the total number of UAV rotor units, and the number of columns is equal to the number of calculated harmonic orders plus one.

[0197] The multi-band dynamic noise reference matrix serves as the reference input signal for the adaptive filter in subsequent steps, providing multi-dimensional noise frequency characteristic information for noise cancellation processing.

[0198] Step S209: The multi-band dynamic noise reference matrix is ​​input as a reference signal into the multi-reference source adaptive filter to track and cancel the aliased wandering aerodynamic noise generated by the differential operation of multiple independent rotors in the original acoustic signal, so as to obtain the target pure acoustic signal.

[0199] Among them, the multi-reference-source adaptive filter refers to an adaptive noise cancellation filter with multiple reference signal input channels. The multi-reference-source adaptive filter can simultaneously receive reference signals from multiple independent noise sources, and perform independent adaptive weight updates and noise estimation for the noise components corresponding to each reference signal. Finally, the noise signals estimated by each reference channel are superimposed and subtracted from the original acoustic signal, thereby achieving synchronous cancellation of mixed noise generated by multiple independent noise sources.

[0200] Aliased drifting aerodynamic noise refers to the composite aerodynamic noise generated by multiple rotor units rotating independently at different and real-time changing speeds during differential hovering. The aerodynamic noise components are superimposed and mixed in space, and the frequency positions of each component drift and float in the frequency domain as the rotation speed of each rotor unit changes in real time.

[0201] Because the rotational speeds of each rotor unit are different and constantly changing, the frequency positions of the aerodynamic noise generated by each rotor unit are also different and constantly drifting. The aerodynamic noise of multiple rotor units forms multiple sets of staggered noise spectrum lines with constantly shifting positions in the frequency domain. These noise spectrum lines are superimposed and mixed at the acoustic sensor array to form aliased and drifting aerodynamic noise.

[0202] In this step, the control unit uses the multi-band dynamic noise reference matrix generated in the previous step as a reference signal input to each reference channel of the multi-reference source adaptive filter. Each reference channel in the multi-reference source adaptive filter corresponds to the noise component of each rotor unit, and each reference channel receives the dynamic blade passage frequency and harmonic frequency information of the corresponding rotor unit as the reference input for that channel.

[0203] The multi-reference source adaptive filter estimates the noise waveform generated by the aerodynamic noise of each rotor unit at the current rotational speed at the acoustic sensor array in real time, based on the input information of each reference channel and through an adaptive algorithm. Specifically, the adaptive algorithm iteratively updates the filter weight coefficients of each reference channel by minimizing the power of the filter output signal, enabling the filter to accurately track the noise frequency shifts caused by real-time changes in the rotational speed of each rotor unit.

[0204] The multi-reference source adaptive filter cancels out the noise waveforms estimated by each reference channel from the original acoustic signal one by one, thereby eliminating the aliased and wandering aerodynamic noise components generated by the differential operation of multiple independent rotors and obtaining the target pure acoustic signal.

[0205] Step S210: Perform a fast Fourier transform on the target pure acoustic signal to obtain the frequency domain signal.

[0206] The Fast Fourier Transform (FFT) is an efficient numerical computation algorithm that converts discrete-time signals into frequency-domain representations. The FFT decomposes the time-domain signal and calculates the amplitude and phase values ​​of each frequency component in the time-domain signal, thereby converting the time-domain signal into a frequency-domain signal with frequency as the independent variable.

[0207] A frequency domain signal is a frequency domain representation obtained after a target pure acoustic signal undergoes a fast Fourier transform. The horizontal axis of a frequency domain signal represents frequency, and the vertical axis represents the amplitude or power spectral density value corresponding to each frequency component.

[0208] In this step, the control unit performs a Fast Fourier Transform (FFT) operation on the target pure acoustic signal obtained in the previous step. The control unit first segments or pads the target pure acoustic signal with zeros according to a preset number of transform points. Then, it performs step-by-step decomposition and frequency domain synthesis operations on the time-domain sampled values ​​of the target pure acoustic signal according to the butterfly operation structure of the FFT, and finally obtains the complex spectrum values ​​of the target pure acoustic signal at each discrete frequency point.

[0209] The control unit performs modulo operations on the complex spectral values ​​at each discrete frequency point to obtain the amplitude value corresponding to each frequency point, thereby forming a frequency domain signal. The frequency domain signal presents the energy contribution of each frequency component in the target pure acoustic signal in an intuitive frequency amplitude distribution form, providing a data foundation for extracting frequency domain feature parameters in subsequent steps.

[0210] Step S211: Extract the main resonant frequency parameter, high-frequency energy attenuation coefficient, and frequency band energy ratio parameter from the frequency domain signal.

[0211] Among them, the main resonant frequency parameter refers to the frequency value corresponding to the frequency component with the largest amplitude in the frequency domain signal. The main resonant frequency parameter characterizes the main vibration frequency of the structure under test after being excited by a mechanical excitation pulse.

[0212] The main resonant frequency parameter of the structure under test is closely related to the local stiffness and mass distribution of the structure under test near the excitation point. When there are damage defects such as cracks or voids inside the structure under test, the damage defects lead to a decrease in local stiffness, and the main resonant frequency parameter will shift downward relative to the normal intact state.

[0213] The high-frequency energy attenuation coefficient refers to the rate attenuation of the spectral amplitude of a frequency domain signal within a preset high-frequency range as the frequency increases. The high-frequency energy attenuation coefficient characterizes the rate of attenuation of high-frequency components in forced radiated sound waves. When damage or defects exist within the structure under test, the scattering and absorption of high-frequency sound waves by these defects are enhanced, leading to an increase in the high-frequency energy attenuation coefficient.

[0214] The band energy ratio parameter refers to the ratio of the integrated energy of a frequency domain signal in two different preset frequency ranges. It characterizes the energy distribution relationship of forced radiated sound waves across different frequency bands. When damage or defects exist within the structure under test, these defects alter the propagation and radiation characteristics of sound waves in different frequency bands, causing the band energy ratio parameter to deviate from its value under normal, intact conditions.

[0215] In this step, the control unit first searches for the frequency peak with the largest amplitude in the frequency domain signal and determines the frequency value corresponding to this peak as the main resonant frequency parameter. Then, within a preset high-frequency band of the frequency domain signal, the control unit performs linear regression fitting or exponential decay fitting on the trend of spectral amplitude change with increasing frequency, and determines the resulting attenuation slope or attenuation constant as the high-frequency energy attenuation coefficient. Finally, the control unit calculates the total integrated energy of the frequency domain signal in the first and second preset frequency bands respectively, and determines the ratio of the total integrated energy in the first preset frequency band to the total integrated energy in the second preset frequency band as the band energy ratio parameter. The frequency band containing the first preset frequency band is higher than the frequency band containing the second preset frequency band.

[0216] Step S212: The main resonant frequency parameter, high-frequency energy attenuation coefficient and frequency band energy ratio parameter are spliced ​​and combined to obtain the target acoustic characteristic parameter set.

[0217] In this step, the control unit combines the three characteristic parameters extracted in the previous step—the main resonant frequency parameter, the high-frequency energy attenuation coefficient, and the frequency band energy ratio parameter—in a preset order to form a target acoustic characteristic parameter set.

[0218] The target acoustic characteristic parameter set consists of three numerical parameters: the main resonant frequency parameter, the high-frequency energy attenuation coefficient, and the frequency band energy ratio parameter. These three parameters quantitatively describe the spectral characteristics of the target pure acoustic signal from three different frequency domain dimensions: the position of the main resonant frequency, the high-frequency attenuation characteristics, and the frequency band energy distribution.

[0219] The target acoustic feature parameter set comprehensively characterizes the acoustic response characteristics of the structure under test at the current target detection position in a synergistic and complementary manner of multi-dimensional features, providing information-rich and dimensionally compact feature inputs for damage identification analysis in subsequent steps.

[0220] Step S213: Construct the target acoustic feature parameter set into a one-dimensional feature vector and input it into the pre-trained structural damage recognition model for feature mapping analysis.

[0221] Among them, a one-dimensional feature vector refers to a one-dimensional numerical array formed by arranging the various feature parameters in the target acoustic feature parameter set in a preset order. The length of the one-dimensional feature vector is equal to the number of feature parameters contained in the target acoustic feature parameter set.

[0222] Specifically, the structural damage identification model can be a classification model based on a neural network architecture, which can map the input one-dimensional feature vector into the probability output of each preset classification category.

[0223] Feature mapping analysis refers to the calculation process by which a structural damage identification model maps the input low-level feature vectors layer by layer to high-level abstract feature representations through weighted summation and nonlinear transformation of neurons in each layer, and finally maps them to the probability distributions of each candidate damage category.

[0224] In this step, the control unit sequentially arranges the main resonant frequency parameter, high-frequency energy attenuation coefficient, and band energy ratio parameter from the target acoustic feature parameter set obtained in the previous step into a one-dimensional feature vector. The control unit then inputs this one-dimensional feature vector into the input layer of the pre-trained structural damage recognition model.

[0225] After receiving a one-dimensional feature vector, the structural damage identification model performs weighted summation and nonlinear activation function transformations on the input data in each hidden layer, achieving a layer-by-layer mapping from the original feature space to the damage discrimination feature space. The output layer of the structural damage identification model further maps the high-level feature representation output by the last hidden layer into numerical values ​​for each candidate damage category, and converts each value into a probability distribution form through a normalized exponential function.

[0226] Step S214: Obtain the anomaly confidence probability values ​​for each candidate damage category output by the structural damage identification model.

[0227] Candidate damage categories refer to a predefined set of various structural state categories that the structural damage identification model can identify and distinguish. Candidate damage categories include no-damage categories and various different damage type categories, such as crack damage, void damage, and erosion damage.

[0228] The anomaly confidence probability value is the numerical value output by the structural damage identification model that represents the probability that an input sample belongs to each candidate damage category. The sum of the anomaly confidence probabilities for each candidate damage category equals one. The higher the anomaly confidence probability value, the higher the degree of confidence the structural damage identification model has in determining that the input sample belongs to that candidate damage category.

[0229] In this step, the control unit reads the output values ​​of each output node of the structural damage identification model after transformation by a normalized exponential function. The output value of each output node is the anomaly confidence probability value for each candidate damage category. The control unit acquires and records the anomaly confidence probability value for each candidate damage category, and determines the anomaly confidence probability value with the largest value and its corresponding candidate damage category, providing a decision-making basis for damage determination in subsequent steps.

[0230] Step S215: If the maximum anomaly confidence probability value is greater than the preset alarm threshold, a damage determination result containing the internal damage type and the estimated damage size is generated.

[0231] The preset alarm threshold refers to a pre-set probability threshold value used to determine whether damage exists. The value of the preset alarm threshold is set according to the actual flaw detection operation's need to balance the risk of missed detection and the risk of false alarm.

[0232] Internal damage type refers to the specific damage type description information corresponding to the candidate damage category corresponding to the highest anomaly confidence probability value. Internal damage type indicates the types of damage that may exist inside the structure under test, such as cracks, cavities, or erosion.

[0233] Predicted damage size refers to the size information of internal damage defects in the structure under test estimated by the structural damage identification model based on the analysis of the input feature vector. The predicted damage size can be obtained through the auxiliary output node of the structural damage identification model or by looking up the mapping relationship table between feature parameter values ​​and damage size established during the training phase.

[0234] In this step, the control unit compares the maximum anomaly confidence probability value determined in the previous step with a preset alarm threshold. When the maximum anomaly confidence probability value is greater than the preset alarm threshold, it indicates that the structural damage identification model determines that the state of the structure under test corresponding to the input target acoustic feature parameter set belongs to a certain damage type category with a confidence level higher than the preset alarm threshold. At this time, the control unit generates a damage determination result.

[0235] The damage assessment result includes two pieces of information: internal damage type and estimated damage size. The internal damage type is a description of the candidate damage category corresponding to the highest anomaly confidence probability value, and the estimated damage size is the estimated size value corresponding to the damage type.

[0236] When the maximum anomaly confidence probability value is less than or equal to the preset alarm threshold, it indicates that the structural damage identification model has not given a high confidence judgment exceeding the preset alarm threshold for any damage type category corresponding to the input target acoustic feature parameter set. In other words, the anomaly confidence probability value for each damage type category has not reached the level required to trigger an alarm. At this time, the control unit generates a normal state determination result. The normal state determination result indicates that the acoustic response characteristics of the structure under test at the current target detection location do not exhibit significant anomalies matching known damage patterns, and no structural damage requiring an alarm is detected at the current target detection location.

[0237] After generating damage assessment results or normal status assessment results, the control unit stores the assessment results in the local storage medium and sends them to the ground monitoring terminal equipment for display output via a wireless communication link.

[0238] On the other hand, if the maximum anomaly confidence probability value is less than or equal to the preset alarm threshold, a normal state judgment result is generated, indicating that no significant anomaly was detected in the structure under test at the current target detection position.

[0239] In some embodiments, step S204 may specifically include the following steps:

[0240] S2041, Obtain the target material type of the structure to be tested.

[0241] The target material type refers to the type of construction material used in the structure under test at the current target detection location. Examples of target material types include, but are not limited to, different categories of building structural materials such as steel, concrete, wood, and stone. Structural materials of different target material types have different physical and mechanical properties, including different parameters such as density, elastic modulus, and sound propagation velocity. These differences in physical and mechanical properties lead to significant differences in the vibration response characteristics of structures of different material types when subjected to the same mechanical excitation pulse.

[0242] In this step, the control unit determines the target material type of the structure under test either through pre-configuration or real-time acquisition.

[0243] The pre-configuration method refers to the operator pre-entering the material type information of the structure to be tested into the task parameter configuration file of the control unit according to the design drawings, construction files or on-site survey results of the structure to be tested before the UAV performs the flaw detection flight mission. When the control unit performs the flaw detection operation, it reads the target material type from the task parameter configuration file.

[0244] The real-time acquisition method refers to the control unit obtaining the target material type by sending material type specification instructions in real time through the remote communication link by the ground operator during the flaw detection operation.

[0245] After the control unit obtains the target material type, it selects the corresponding excitation parameter combination according to the specific category of the target material type to control the electromagnetic catapult component to output mechanical excitation pulses that are compatible with the target material type.

[0246] S2042, if the target material type is the first material, then the electromagnetic catapult assembly is controlled to output a first mechanical excitation pulse to the structure under test with a first driving current and a first pulse width. If the target material type is the second material, then the electromagnetic catapult assembly is controlled to output a second mechanical excitation pulse to the structure under test with a second driving current and a second pulse width.

[0247] In this context, the first material and the second material represent two structural material categories with different acoustic impedances. Acoustic impedance is a physical parameter characterizing the degree to which a material impedes the propagation of sound waves; it is equal to the product of the material's density and the speed of sound propagation within that material. Materials with higher acoustic impedance exhibit a smaller mechanical response to external excitation, requiring greater excitation energy to generate forced radiation sound waves that meet detection requirements. The first material has a higher acoustic impedance than the second material. In a specific example, the first material could be steel, and the second material could be wood. Steel has a higher density and sound velocity, resulting in a much higher acoustic impedance than wood.

[0248] The first driving current refers to the amplitude of the pulse driving current applied to the electromagnetic coil of the electromagnetic catapult assembly when the target material type is the first material. The second driving current refers to the amplitude of the pulse driving current applied to the electromagnetic coil of the electromagnetic catapult assembly when the target material type is the second material. The first driving current is greater than the second driving current. The amplitude of the driving current determines the strength of the pulsed magnetic field generated by the electromagnetic coil, which in turn determines the magnitude of the electromagnetic driving force and the impact velocity of the catapult impactor. The larger the driving current, the higher the impact velocity of the catapult impactor and the greater the peak force of the output mechanical excitation pulse.

[0249] Because the acoustic impedance of the first material is relatively large, a large first driving current is required to drive the electromagnetic catapult assembly and output a first mechanical excitation pulse with a large peak force in order to excite a forced vibration response with sufficient amplitude in the structure under test of the first material.

[0250] The first pulse width refers to the duration of the drive current pulse applied to the electromagnetic coil when the target material type is the first material. The second pulse width refers to the duration of the drive current pulse applied to the electromagnetic coil when the target material type is the second material. The first pulse width is shorter than the second pulse width.

[0251] The pulse width determines the duration of the mechanical excitation pulse acting on the surface of the structure under test. The shorter the pulse width, the wider the frequency bandwidth excited by the excitation pulse. The longer the pulse width, the narrower the frequency bandwidth excited by the excitation pulse and the more concentrated it is in the lower frequency band.

[0252] For a first material with a high acoustic impedance, since the first material usually has high stiffness and high natural frequency, a shorter first pulse width can be used to excite broadband forced radiation acoustic waves containing high frequency components, which is beneficial for detecting small damage defects in the first material structure.

[0253] For a second material with low acoustic impedance, since the second material usually has low stiffness and low natural frequency, using a longer second pulse width can concentrate the excitation energy in the lower frequency band where the second material structure is prone to resonant response, which is beneficial to excite forced radiation sound waves with sufficient amplitude.

[0254] The first mechanical excitation pulse refers to the mechanical excitation pulse output by the electromagnetic catapult assembly under the control parameters of the first driving current and the first pulse width. The first mechanical excitation pulse has a high peak force and a short pulse duration.

[0255] The second mechanical excitation pulse refers to the mechanical excitation pulse output by the electromagnetic catapult assembly under the control parameters of the second drive current and the second pulse width. The second mechanical excitation pulse has a relatively low peak force and a long pulse duration.

[0256] In this step, the control unit makes a conditional judgment based on the target material type obtained in step S2041.

[0257] When the target material type is the first material, the control unit sends an excitation control command with the first driving current and the first pulse width as parameters to the drive circuit of the electromagnetic catapult assembly. The drive circuit generates a corresponding pulse driving current according to the received first driving current amplitude value and first pulse width duration value and applies it to the electromagnetic coil. Under the drive of the first driving current and the first pulse width, the electromagnetic catapult assembly outputs the first mechanical excitation pulse. The first mechanical excitation pulse acts on the surface of the structure to be tested through the flexible buffer mechanism.

[0258] When the target material type is the second material, the control unit sends an excitation control command with the second driving current and the second pulse width as parameters to the drive circuit of the electromagnetic catapult assembly. The drive circuit generates a corresponding pulse driving current according to the received second driving current amplitude value and second pulse width duration value and applies it to the electromagnetic coil. Under the drive of the second driving current and the second pulse width, the electromagnetic catapult assembly outputs a second mechanical excitation pulse. The second mechanical excitation pulse acts on the surface of the structure to be tested through a flexible buffer mechanism.

[0259] This embodiment adopts an adaptive matching method of excitation parameters based on the target material type, so that the electromagnetic catapult component outputs mechanical excitation pulses that are compatible with the material characteristics of the structure under test.

[0260] Specifically, the material adaptive excitation parameter matching mechanism implemented by the control unit through the above steps S2041 and S2042 enables the controllable excitation device to automatically adjust the excitation output parameters according to the different target material types of the structure under test, thereby outputting mechanical excitation pulses that are compatible with structural materials with different acoustic impedance characteristics.

[0261] For a first material structure with a high acoustic impedance, a larger first driving current is used to provide sufficient excitation energy to overcome the high acoustic impedance characteristics of the first material, while a shorter first pulse width is used to excite a broadband frequency response suitable for the high stiffness characteristics of the first material.

[0262] For a second material structure with low acoustic impedance, a smaller second driving current is used to avoid excessive excitation force from damaging the surface of the second material structure. At the same time, a longer second pulse width is used to concentrate the excitation energy in the lower frequency band where the second material is more likely to produce an effective vibration response.

[0263] By using the material-adaptive excitation parameter matching method described above, the UAV can generate forced radiated sound waves that meet the requirements of acoustic flaw detection when facing test structures of different material types, thus improving the applicability of this method to building components of different material types.

[0264] In some embodiments, step S211 may specifically include the following steps:

[0265] S2111, extract the first integral total value of the frequency domain signal within the first preset frequency range.

[0266] The first preset frequency interval refers to a continuous frequency range predefined on the frequency axis of the frequency domain signal. The first preset frequency interval is defined by a pre-set lower frequency limit and a first upper frequency limit. The first integrated total energy value refers to the cumulative sum of the energy of all frequency components contained within the first preset frequency interval of the frequency domain signal. The first integrated total energy value is obtained by integrating the power spectral density value of the frequency domain signal along the frequency axis within the first preset frequency interval. In the case of discrete frequency domain signals, the first integrated total energy value is calculated by summing the squares of the amplitude values ​​at each discrete frequency point in the frequency domain signal that falls within the first preset frequency interval.

[0267] In this step, the control unit first determines the first lower frequency limit and the first upper frequency limit of the first preset frequency range. The control unit locates all discrete frequency points in the frequency domain signal obtained in step S210 whose frequency values ​​fall between the first lower frequency limit and the first upper frequency limit. It extracts the corresponding spectral amplitude values ​​at these discrete frequency points, squares each spectral amplitude value, and then sums them point by point. The summation result is determined as the first total integrated energy value. The first total integrated energy value represents the total energy of the frequency domain signal within the first preset frequency range in the form of a single numerical value.

[0268] Step S2112: Extract the second integral total value of the frequency domain signal within the second preset frequency range.

[0269] The frequency band of the first preset frequency range is higher than the frequency band of the second preset frequency range.

[0270] The second preset frequency range refers to another continuous frequency range predefined on the frequency axis of the frequency domain signal. The second preset frequency range is defined by a pre-set second lower frequency limit and a second upper frequency limit. The second preset frequency range does not overlap with the first preset frequency range on the frequency axis, and the frequency band of the first preset frequency range is higher than the frequency band of the second preset frequency range, that is, the first lower frequency limit is greater than the second upper frequency limit.

[0271] The second integral total energy value refers to the cumulative sum of the energy of all frequency components contained in the frequency domain signal within the second preset frequency interval. The calculation method of the second integral total energy value is the same as that of the first integral total energy value, which is to sum the squares of the amplitude values ​​of each discrete frequency point in the frequency domain signal that falls within the second preset frequency interval point by point.

[0272] In this step, the control unit determines the second lower frequency limit and the second upper frequency limit of the second preset frequency range. The control unit locates all discrete frequency points in the frequency domain signal whose frequency values ​​fall between the second lower frequency limit and the second upper frequency limit, extracts the corresponding spectral amplitude values ​​at these discrete frequency points, squares each spectral amplitude value, and then sums them point by point. The summation result is determined as the second total integrated energy value. Since the frequency band of the first preset frequency range is higher than the frequency band of the second preset frequency range, the first total integrated energy value reflects the energy magnitude of the frequency domain signal in the higher frequency band, while the second total integrated energy value reflects the energy magnitude of the frequency domain signal in the lower frequency band.

[0273] In practical applications, the specific frequency ranges of the first and second preset frequency ranges are set according to the material type of the structure under test and the spectral variation law corresponding to common damage modes, so that the energy comparison relationship between the two frequency ranges has sensitive response characteristics to the existence of structural damage.

[0274] Step S2113: Calculate the ratio of the first integral total energy value to the second integral total energy value to obtain the frequency band energy ratio parameter.

[0275] In this step, the control unit uses the first total integral energy value obtained in step S301 as the numerator and the second total integral energy value obtained in step S302 as the denominator, performs a division operation, and determines the quotient obtained by the division operation as the frequency band energy ratio parameter.

[0276] The frequency band energy ratio parameter, in the form of a dimensionless ratio, characterizes the energy distribution relationship between the frequency domain signal in the first preset frequency range and the second preset frequency range, that is, the relative magnitude relationship between the integral energy of the higher frequency band and the integral energy of the lower frequency band.

[0277] When the internal structure of the structure under test is in a normal and intact state, the energy of the forced radiated sound wave exhibits a specific distribution ratio corresponding to the normal and intact state between the first preset frequency range and the second preset frequency range, and the frequency band energy ratio parameter is within the normal value range corresponding to the normal and intact state.

[0278] When damage defects such as cracks, voids, or erosion exist inside the structure under test, these defects alter the propagation and radiation characteristics of sound waves within the structure, causing a shift in the energy distribution ratio of forced radiated sound waves between high and low frequencies. For example, the presence of damage defects such as cracks or voids may enhance the scattering and absorption of high-frequency components, resulting in a decrease in the integrated energy in the high-frequency band relative to the low-frequency band, and a corresponding decrease in the value of the band energy ratio parameter.

[0279] The frequency band energy ratio parameter adopts the energy ratio of two frequency bands. Compared with the absolute energy value of a single frequency band, the ratio form has the self-normalization characteristic of excitation force fluctuation and propagation distance variation, which can reduce the impact of excitation consistency deviation and acquisition distance difference on the stability of characteristic parameters.

[0280] The method of this embodiment will be further described in detail below with reference to two typical application scenarios.

[0281] Scenario 1: Detection of internal cracks in the weld seams of a standard tower crane section.

[0282] This embodiment describes an application scenario for detecting internal cracks in the butt weld of the main chord member of a standard tower crane section. The structure under test is the butt weld area of ​​the main chord member of the standard tower crane section. This butt weld is suspected of having an internal crack at the root due to lack of fusion, with a crack width of approximately 0.1 mm and a crack length of approximately 15 mm. This weld is located at a high altitude within the standard tower crane section, making it difficult for personnel to directly access it for close-range manual flaw detection.

[0283] The UAV used in this embodiment is a six-axis industrial UAV equipped with a real-time dynamic differential positioning module, achieving a positioning accuracy of ±1 cm. The acoustic sensor array employs a 7-element spiral array configuration with a frequency response range of 20 Hz to 20 kHz and an array directivity angle of ±15 degrees. The acoustic sensor array incorporates a field-programmable gate array (FPGA) chip for real-time signal preprocessing. The electromagnetic catapult component within the controllable excitation device operates at a frequency range of 50 Hz to 5 kHz. The control unit utilizes an edge computing module and is pre-loaded with a structural damage recognition model trained for steel structure damage scenarios.

[0284] The specific implementation process of this embodiment is as follows.

[0285] The control unit guides the drone to fly to the area where the target weld is located, identifies the weld trajectory position through an onboard vision sensor, and controls the drone to hover at the target detection position approximately 3 centimeters above the weld surface. After reaching the target detection position, the drone maintains stable hovering with centimeter-level accuracy using a real-time dynamic differential positioning module and a vision positioning sensor.

[0286] The control unit guides the UAV to continue approaching the structure under test until the flexible buffer mechanism at the front end of the controllable excitation device contacts the weld surface of the structure under test. The control unit acquires real-time compression data of the flexible buffer mechanism. When the real-time compression data reaches the preset buffer threshold, the control unit controls the UAV to output anti-recoil compensation thrust using the flight control system to maintain the contact state of the flexible buffer mechanism and the current hovering attitude.

[0287] Since the target material of the structure under test is steel, which is a material with relatively high acoustic impedance, the control unit controls the electromagnetic catapult assembly to output a first mechanical excitation pulse to the structure under test with a first driving current and a first pulse width. In this embodiment, the frequency range of the excitation signal emitted by the electromagnetic catapult assembly covers 500Hz to 3000Hz, the excitation period is 2 seconds, and the excitation operation is repeated 3 times to excite the weld and the surrounding metal area to generate micro-amplitude forced vibration. Because steel has a relatively high acoustic impedance, using a larger first driving current can output a first mechanical excitation pulse with a higher peak force. At the same time, using a shorter first pulse width ensures that the excited forced radiated sound wave contains rich high-frequency components, which is beneficial for detecting fine crack defects with a width of about 0.1mm inside the weld.

[0288] The acoustic sensor array synchronously acquires raw acoustic signals, including forced radiated sound waves and UAV platform noise, at a sampling rate of 96kHz. The control unit synchronously acquires the independent real-time rotational speed data of each rotor unit while the UAV is in differential hovering state.

[0289] The control unit calculates the dynamic blade passage frequency and its harmonic frequencies for each rotor unit based on the real-time rotational speed data and the number of blades configured for each rotor unit. In this embodiment, the main energy of the UAV platform noise is concentrated in the 200Hz to 400Hz range and its integer harmonic frequencies. The control unit aggregates the dynamic blade passage frequencies and their harmonic frequencies corresponding to each rotor unit to generate a multi-band dynamic noise reference matrix. This multi-band dynamic noise reference matrix is ​​then input as a reference signal into a multi-reference source adaptive filter to track and cancel the aliased aerodynamic noise generated by the differential operation of multiple independent rotors in the original acoustic signal. After noise cancellation processing, the signal-to-noise ratio of the output target clean acoustic signal is improved by approximately 20dB compared to the original acoustic signal.

[0290] The control unit performs a Fast Fourier Transform on the target pure acoustic signal to obtain a frequency domain signal. The main resonant frequency parameter, high-frequency energy attenuation coefficient, and bandwidth energy ratio parameter are extracted from the frequency domain signal, and these three characteristic parameters are concatenated to obtain the target acoustic characteristic parameter set.

[0291] In this embodiment, for the butt weld of the main chord of the standard section of this type of tower crane, under a healthy state with no internal damage, the baseline values ​​of various characteristic parameters obtained through the same excitation and acquisition process are as follows: The healthy baseline value for the main resonant frequency parameter is 1850Hz; the healthy baseline value for the high-frequency energy attenuation coefficient is less than 1dB per octave; and the frequency band energy ratio parameter is within the normal range. These healthy baseline values ​​were obtained statistically from a large number of data collections of similar healthy weld specimens during the training phase of the structural damage identification model.

[0292] The specific values ​​of each characteristic parameter in the target acoustic characteristic parameter set obtained in this embodiment are as follows. The measured value of the main resonant frequency parameter is 1700Hz, which is about 8% lower than the healthy reference value of 1850Hz. This shift indicates that the crack defects inside the weld cause a decrease in the local stiffness of the weld area, thus causing the main resonant frequency parameter to drift towards lower frequencies. The measured value of the high-frequency energy attenuation coefficient reaches 8dB in the 2500Hz to 2800Hz frequency band, which is significantly greater than the healthy reference value of less than 1dB per octave. This increase indicates that the crack interface inside the weld has a strong scattering and absorption effect on high-frequency sound waves, resulting in a significant increase in the attenuation of high-frequency energy.

[0293] The control unit constructs a one-dimensional feature vector from the target acoustic feature parameter set and inputs it into a pre-trained structural damage recognition model for feature mapping analysis. The structural damage recognition model outputs an anomaly confidence probability value of 93% for the weld internal crack damage category, which is the highest anomaly confidence probability value among all candidate damage categories, and this value is greater than the preset alarm threshold. Based on this, the control unit generates a damage determination result, which includes the internal damage type as weld internal crack, with an estimated damage size of approximately 1.5 mm to 2.0 mm in crack depth.

[0294] The control unit sends the damage assessment results to the ground-based monitoring terminal for display and output. The damaged location of the weld is marked in red on the three-dimensional point cloud model, and an inspection report is generated that includes a comparison chart of the measured spectrum and the health baseline spectrum. The inspection report also includes maintenance recommendations to re-inspect using ultrasonic methods.

[0295] Scenario 2: Detection of termite-infested beams and columns in the wooden structure of ancient buildings.

[0296] This embodiment is applied to the detection of termite infestation in the fir wood pillars of a Qing Dynasty wooden-structured residence. The structure under test is a fir wood pillar with a diameter of approximately 300mm. The outer surface of the pillar shows no visible damage, but the sound produced when manually tapped is abnormal, suggesting the presence of termite-eaten cavities inside the pillar. This fir wood pillar is located indoors and is a structural component of a protected historical building.

[0297] The drone used in this embodiment is a lightweight quadcopter. Since the detection environment is indoors and satellite positioning is not available, the drone uses visual positioning to achieve hovering and positioning within the indoor space. The acoustic sensor array adopts a 5-element ring array configuration with a frequency response range of 20Hz to 10kHz and an array directivity angle of ±20 degrees. The controllable vibration device is a contact-type configuration, and the flexible buffer mechanism at the front end of the controllable vibration device uses a rubber buffer head. The control unit uses an embedded computing module paired with a dedicated neural network acceleration chip, pre-loaded with a structural damage recognition model trained for wood damage scenarios.

[0298] The specific implementation process of this embodiment is as follows.

[0299] The control unit controls the drone to fly to the target detection position at a height of about 1.5m above the cedar pillar, and uses a visual positioning sensor to achieve stable hovering in an indoor environment.

[0300] The control unit guides the UAV to approach the structure under test until the flexible buffer mechanism at the front end of the controllable excitation device lightly touches the surface of the cedar pillar. After the rubber buffer head of the flexible buffer mechanism forms contact with the surface of the cedar pillar, the control unit acquires real-time compression data of the flexible buffer mechanism. When the real-time compression data reaches a preset buffer threshold, the control unit controls the flight control system to output anti-recoil compensation thrust to maintain the contact state of the flexible buffer mechanism and the current hovering attitude. In this embodiment, the pre-pressure formed after the flexible buffer mechanism contacts the surface of the cedar pillar is approximately 0.5N. This pre-pressure value ensures that the mechanical excitation pulse output by the electromagnetic catapult assembly is reliably transmitted to the surface of the cedar pillar through the flexible buffer mechanism without damaging the surface of the wooden artifact structure.

[0301] Since the target material of the structure under test is wood, which is a material with low acoustic impedance, the control unit controls the electromagnetic catapult assembly to output a second mechanical excitation pulse to the structure under test with a second driving current and a second pulse width. In this embodiment, the frequency range of the excitation signal emitted by the electromagnetic catapult assembly covers 100Hz to 800Hz, and the excitation period is 3 seconds. Because wood has low acoustic impedance and the natural frequency of the cedar column is low, a smaller second driving current is used to avoid mechanical damage to the surface of the wooden artifact. At the same time, a longer second pulse width is used to concentrate the excitation energy in the low-frequency band where wood easily produces effective bending vibration mode response, so as to excite the cedar column to generate forced radiated sound waves containing information about its internal structural state.

[0302] An acoustic sensor array synchronously acquires raw acoustic signals containing forced radiated sound waves and UAV platform noise. The control unit obtains the independent real-time rotational speed data of each rotor unit when the UAV is in differential hovering state. Based on the real-time rotational speed data of each rotor unit and the number of blades configured for each rotor unit, the control unit calculates the dynamic blade passage frequency and its harmonic frequencies for each rotor unit. In this embodiment, since the UAV is a quadcopter, the main harmonic components of the UAV platform noise of each rotor unit are concentrated around 400Hz and 800Hz.

[0303] The control unit aggregates the dynamic blades corresponding to each rotor unit through their frequencies and harmonic frequencies to generate a multi-band dynamic noise reference matrix. The multi-band dynamic noise reference matrix is ​​then input as a reference signal into a multi-reference source adaptive filter to track and cancel the aliased wandering aerodynamic noise generated by the differential operation of multiple independent rotors in the original acoustic signal, thereby obtaining the target pure acoustic signal.

[0304] The control unit performs a Fast Fourier Transform on the target pure acoustic signal to obtain a frequency domain signal. From the frequency domain signal, the main resonant frequency parameter, the high-frequency energy attenuation coefficient, and the bandgap energy ratio parameter are extracted.

[0305] In this embodiment, the band energy ratio parameter is calculated by extracting the first total integral energy value of the frequency domain signal within a first preset frequency range and the second total integral energy value within a second preset frequency range, and then calculating the ratio of the first total integral energy value to the second total integral energy value. The first preset frequency range is the mid-frequency band from 200Hz to 500Hz, and the second preset frequency range is the low-frequency band from 50Hz to 200Hz. The frequency band of the first preset frequency range is higher than that of the second preset frequency range. The band energy ratio parameter reflects the relative distribution relationship between the mid-frequency band energy and the low-frequency band energy. When termite-eaten cavities exist inside the fir wood column, the cavities will produce a local resonance effect, causing the integrated energy of the mid-frequency band to rise abnormally relative to the low-frequency band, and the value of the band energy ratio parameter will increase accordingly.

[0306] The control unit splices and combines the main resonant frequency parameters, the high-frequency energy attenuation coefficient, and the frequency band energy ratio parameters to obtain the target acoustic characteristic parameter set.

[0307] In this embodiment, for a 300mm diameter cedar post, under healthy conditions with no internal damage, the health baseline values ​​of various characteristic parameters obtained through the same excitation and acquisition process are as follows: the health baseline value for the main resonant frequency parameter is 320Hz, the health baseline value for the bandwidth energy ratio parameter is 0.40, and the health baseline value for the pulse decay time is 200 milliseconds.

[0308] The specific values ​​of each characteristic parameter in the target acoustic characteristic parameter set obtained in this embodiment are as follows: The measured value of the main resonant frequency parameter is 275Hz, which is about 14% lower than the healthy reference value of 320Hz. The physical reason for the lower frequency shift of the main resonant frequency parameter is that termite infestation has caused local material loss inside the fir column, reducing the effective load-bearing cross section and decreasing the local stiffness, thus lowering the main vibration frequency of the fir column. The measured value of the frequency band energy ratio parameter is 0.58, which is about 45% higher than the healthy reference value of 0.40.

[0309] The physical reason for the increased frequency band energy ratio parameter lies in the fact that the hollow cavities inside the cedar wood pillars form a local resonant space. The natural frequency of this resonant space falls within the mid-frequency range of 200Hz to 500Hz, causing an abnormal enhancement of the mid-frequency energy in the forced radiated sound waves, resulting in a significant increase in the energy ratio of the mid-frequency to low-frequency bands. The measured value of the pulse decay time associated with the high-frequency energy attenuation coefficient is 110 milliseconds, which is about 45% shorter than the healthy baseline value of 200 milliseconds. The physical reason for the shortened decay time is that the rough interface of the hollow cavities generates a frictional energy dissipation effect on the sound waves, accelerating the attenuation and dissipation of the sound wave vibration energy.

[0310] The control unit constructs a one-dimensional feature vector from the target acoustic feature parameter set and inputs it into a pre-trained structural damage identification model for feature mapping analysis. The structural damage identification model outputs an anomaly confidence probability value of 94% for the internal termite-borne cavity damage category in the wood structure, which is the highest anomaly confidence probability value among all candidate damage categories, and this value is greater than the preset alarm threshold. Based on this, the control unit generates a damage determination result, which includes the internal damage type as internal termite-borne cavity, with an estimated damage size of approximately 450 cm³. 3 Up to 550cm 3 The depth of the borehole is about 20mm to 30mm.

[0311] The control unit sends the damage assessment results to the ground-based monitoring terminal for display and output. The location of the borer-damaged cedar post is marked in orange on the 3D model, and a detection report is generated. The report includes a comparison table of measured spectral characteristic parameters and healthy baseline characteristic parameters, as well as repair recommendations for reinforcement using resin injection.

[0312] Referring to Table 1, the health baseline value, measured value, variation range, and corresponding physical meaning of each feature parameter in this scenario embodiment are listed.

[0313] <![CDATA[F0]]> 320 Hz 275 Hz -14% Hollowing leads to a decrease in local stiffness R (Medium / Low Frequency Energy Ratio) 0.40 0.58 +45% Local resonance occurs in the cavity of the borer. τ (decay time) 200 ms 110 ms -45% Energy loss due to friction at rough surfaces in tunnels

[0314] Table 1

[0315] The healthy baseline value for the main resonant frequency parameter is 320Hz, while the measured value is 275Hz, representing a decrease of 14%. The corresponding physical meaning is that termite infestation caused a decrease in the local stiffness of the fir column, resulting in a shift of the main vibration frequency to a lower frequency.

[0316] The healthy baseline value for the frequency band energy ratio parameter is 0.40, and the measured value is 0.58, with a change of 45%. The corresponding physical meaning is that the cavity generates a local resonance effect, which causes the radiated sound wave energy in the mid-frequency band to be abnormally enhanced compared to the low-frequency band.

[0317] The healthy baseline value for pulse decay time is 200 milliseconds, while the measured value is 110 milliseconds, representing a 45% reduction. The corresponding physical meaning is that the frictional energy dissipation effect of the rough interface of the cavity accelerates the decay of acoustic vibration energy.

[0318] As can be seen from the two specific application scenarios above, the method of this embodiment can adaptively match the excitation parameters for the test structure of different material types, and effectively identify hidden damage inside the test structure by extracting acoustic spectrum features from the target pure acoustic signal and reasoning and analyzing the structural damage identification model. This includes damage types that cannot be detected by traditional visual inspection methods, such as internal cracks in steel structure welds and termite infestation in wooden structures.

[0319] In other application scenarios, when performing flaw detection on non-flat surfaces such as wind turbine blades, complex pipelines, and ancient buildings with irregularly shaped decorative components, the drone may be affected by crosswinds or airflow disturbances in the air, making it difficult to ensure that the axis of the controllable excitation device is strictly perpendicular to the local tangential plane of the structure under test.

[0320] When the flexible buffer mechanism contacts the surface of the structure under test at a certain angle, the force at the contact interface exhibits an asymmetrical distribution. On one hand, this causes the excitation force to decompose into normal and tangential components. The tangential component will excite strong shear stress waves and surface waves inside the structure. These unexpected wave types will overlap with longitudinal waves reflecting internal damage, causing severe distortion of the forced radiated sound wave spectrum. On the other hand, the geometric misfit at the microscopic level will cause abrupt changes in acoustic impedance at the contact interface, resulting in a large amount of excitation energy being reflected and unable to be effectively injected into the interior of the structure under test.

[0321] Therefore, this embodiment addresses the application scenario where the surface of the structure under test exhibits a non-flat curved surface. It improves the flexible buffer mechanism in the aforementioned embodiment by introducing a material phase transformation control step into the excitation coupling process. The method of this embodiment further includes the following steps S301 to S305.

[0322] S301, Determine the conformal phase change buffer mechanism.

[0323] Among them, the conformal phase change buffer mechanism refers to a contact coupling component installed at the front end of the controllable excitation device, which is composed of a magnetorheological elastomer matrix material and a multi-pole electromagnetic excitation array arranged around it, and has the ability to switch between controllable stiffness states.

[0324] Magnetorheological elastomer (MRE) matrix materials are composite elastic materials made by uniformly dispersing magnetic microparticles in a silicone rubber matrix according to a predetermined volume fraction and then cross-linking and curing them under an applied directional magnetic field. MRE matrix materials exhibit a low elastic modulus and ultra-flexible state when no external magnetic field is applied. When an external magnetic field is applied, the internal magnetic microparticles align along the magnetic field lines to form a particle chain structure, significantly increasing the overall elastic modulus within milliseconds and exhibiting a high-stiffness near-solid state.

[0325] A multi-pole electromagnetic excitation array refers to an array structure composed of multiple independent and controllable miniature electromagnetic coils arranged at a preset pole spacing along the outer periphery of a magnetorheological elastomer matrix material. When energized, the multi-pole electromagnetic excitation array can generate a magnetic field with a preset distribution pattern in the internal space of the magnetorheological elastomer matrix material.

[0326] S302, Flexible bonding stage control.

[0327] Specifically, during the phase where the control unit controls the UAV to approach and contact the structure under test, the multi-pole electromagnetic excitation array is kept de-energized. At this time, the magnetorheological elastomer matrix material is in an ultra-flexible state, capable of generating large-scale plastic deformation under the thrust of the UAV, actively conforming to the macroscopic curvature and microscopic unevenness of the surface of the structure under test, so that the contact end face of the conformal phase change buffer mechanism forms a large-area seamless and tight contact with the surface of the structure under test.

[0328] S303, Adhesion status detection.

[0329] Specifically, the control unit uses a contact pressure distribution sensor array located inside the conformal phase change buffer mechanism to detect the contact pressure distribution between the conformal phase change buffer mechanism and the surface of the structure under test in real time. The contact pressure distribution sensor array consists of multiple miniature pressure sensing units embedded near the contact end face of the magnetorheological elastomer matrix material in a preset two-dimensional grid array. When the contact pressure distribution sensor array detects that the pressure values ​​of each miniature pressure sensing unit on the contact end face exceed the preset minimum contact pressure threshold, the control unit determines that the contact state has reached stability.

[0330] S304, magneto-induced phase change curing.

[0331] Specifically, after determining that the bonding state is stable, the control unit sends a phase change excitation command to the multi-pole electromagnetic excitation array. Each miniature electromagnetic coil in the multi-pole electromagnetic excitation array is independently energized according to the pressure distribution data fed back by the contact pressure distribution sensor array, applying a stronger magnetic field to areas with higher contact pressure and a weaker magnetic field to areas with lower contact pressure. Under the non-uniform magnetic field of the multi-pole electromagnetic excitation array, the magnetorheological elastomer matrix material undergoes a phase change within milliseconds, transforming from an ultra-flexible state to a high-stiffness solid state. Furthermore, the solidified geometry maintains consistency with the contour formed when the material is bonded to the surface of the structure under test, thus forming a shaped solid vibration anvil whose end face highly matches the surface shape of the structure under test.

[0332] S305, excitation and magneto-reset.

[0333] Specifically, after the conformal phase change buffer mechanism completes the magnetostrictive phase change curing, the control unit controls the electromagnetic catapult assembly to output mechanical excitation pulses in accordance with the manner described in the aforementioned embodiment. The mechanical excitation pulses are uniformly transmitted into the interior of the structure under test through a high-rigidity shaped solid anvil in a surface-contact manner along a direction perpendicular to the local cross-section of the structure under test.

[0334] After the excitation pulse is output, the control unit sends a reverse demagnetizing pulse sequence to the multi-pole electromagnetic excitation array to eliminate the residual magnetization effect in the magnetorheological elastomer matrix material and cuts off the power supply to the multi-pole electromagnetic excitation array. The magnetorheological elastomer matrix material returns to its ultra-flexible state, and the control unit controls the UAV to retreat away from the surface of the structure under test.

[0335] This embodiment breaks through the traditional design concept that the buffer mechanism must be a passive elastic body with a single stiffness. It introduces the magnetostrictive phase transition mechanism of smart materials into the construction process of the air-excited coupling interface. Through a two-stage working mode of first flexible bonding and then rigid curing, it solves the dual contradiction of low excitation energy coupling efficiency and insufficient attitude control accuracy of UAV under non-flat interface.

[0336] In some embodiments, although the additive noise generated by the UAV rotor can be processed by a multi-reference source adaptive filter, in the special scenario of hovering near the wall, the high-speed downwash airflow generated by the UAV rotor will continuously impact the surface of the structure under test, and form a strongly turbulent aerodynamic boundary layer with dynamically changing thickness and dense vortices near the surface of the structure.

[0337] Forced-radiated sound waves, after radiating from the structure's surface into the air, must penetrate this highly turbulent boundary layer to reach the UAV's acoustic sensor array. The drastic random fluctuations in air density and pressure within the turbulent medium cause severe refraction, phase modulation, and broadband scattering of the sound waves during propagation. This constitutes multiplicative channel interference, disrupting the high-frequency phase coherence of the acoustic signal and causing random drift in characteristic parameters such as the high-frequency energy attenuation coefficient and the main resonant frequency. Conventional additive noise cancellation algorithms based on rotational speed are ill-suited to handling the signal distortion caused by the random changes in the physical properties of the sound wave propagation medium itself.

[0338] Therefore, this embodiment addresses the application scenario where, during drone hovering operations near a wall, the strong turbulent boundary layer formed by the rotor downwash airflow impacting the wall causes phase modulation and broadband scattering of forced radiated sound waves. It introduces an active aerodynamic medium control step into the acoustic signal acquisition process. Specifically, this includes the following steps S401 to S404.

[0339] S401 provides a laminar flow generation array.

[0340] Among them, the laminar flow generation array refers to the aerodynamic actuator that surrounds the acoustic sensor array of the UAV and is composed of multiple miniature high-pressure jet generators and rectifier grids corresponding to each miniature high-pressure jet generator.

[0341] A miniature high-pressure jet generator is a small airflow generating device capable of ejecting compressed air at high speed from a miniature nozzle. A flow straightener is a structure composed of multiple parallel thin-walled guide vanes arranged at equal intervals. The function of the flow straightener is to straighten and regulate the initial airflow ejected by the miniature high-pressure jet generator into a laminar airflow with parallel and consistent flow direction and a low Reynolds number.

[0342] The micro high-pressure jet generators in the laminar flow generation array are uniformly and symmetrically arranged along the circumference of the acoustic sensor array, so that the reaction thrust generated by each micro high-pressure jet generator cancels each other out at the center of the UAV fuselage.

[0343] S402, real-time turbulence intensity detection.

[0344] Specifically, the control unit uses a miniature turbulence intensity sensor located near the acoustic sensor array to detect the turbulence intensity value in the airflow space between the current acoustic sensor array and the surface of the structure under test in real time.

[0345] The miniature turbulence intensity sensor is implemented using a hot-wire anemometer or a microcomputer voltage difference sensor, and can output real-time values ​​characterizing the intensity of turbulence fluctuations in the current airflow space.

[0346] S403, synchronously triggers laminar flow jetting.

[0347] Within a preset lead time window before the controllable excitation device outputs mechanical excitation pulses, the control unit sends a laminar flow jet trigger command to the laminar flow generation array based on the real-time turbulence intensity value fed back by the micro turbulence intensity sensor.

[0348] The jet velocity parameter included in the laminar jet trigger command is positively correlated with the real-time turbulence intensity value. That is, the higher the detected turbulence intensity value, the greater the jet velocity indicated by the laminar jet trigger command.

[0349] Upon receiving the laminar jet trigger command, the laminar flow generation array synchronously activates each miniature high-pressure jet generator, ejecting a ring-shaped converging laminar airflow into the space between the acoustic sensor array and the surface of the structure under test. The dynamic head of this ring-shaped converging laminar airflow is higher than that of the rotor downwash airflow in this space. Leveraging its momentum advantage, it pushes the irregular rotor downwash turbulence outwards, creating a cylindrical low-turbulence gas isolation zone within the originally turbulent boundary layer space. This cylindrical low-turbulence gas isolation zone extends from the receiving surface of the acoustic sensor array to the target detection location on the surface of the structure under test, forming a physical air waveguide tunnel with a stable air density gradient and a stable velocity distribution.

[0350] S404, time-window synchronized excitation and acquisition.

[0351] Specifically, after the physical air waveguide tunnel is formed and stabilized, the control unit triggers the controllable excitation device to output mechanical excitation pulses, and simultaneously starts the acoustic sensor array to perform acoustic signal acquisition within a preset acquisition time window.

[0352] Forced radiated sound waves are emitted from the surface of the structure under test and propagate upwards along the physical air waveguide tunnel to the acoustic sensor array. The low turbulence characteristics of the airflow inside the physical air waveguide tunnel prevent significant phase modulation and broadband scattering of the forced radiated sound waves during propagation, thus preserving the phase coherence and spectral integrity of the forced radiated sound waves in the high-frequency band.

[0353] The degree of multiplicative channel interference caused by the rotor downwash airflow in the raw acoustic signal acquired by the acoustic sensor array is significantly reduced. After the preset acquisition time window ends, the control unit immediately shuts down the laminar flow generator array to avoid unnecessary energy consumption of the UAV due to continuous jetting.

[0354] This embodiment departs from the traditional approach of repairing damaged acoustic data at the signal processing algorithm level. Instead, it takes a novel approach by actively controlling the physical properties of the sound wave propagation medium. By constructing a laminar gas isolation zone with low turbulence, it reconstructs the acoustic transmission channel at the physical level, thereby improving the acoustic capture quality of UAVs in extreme airflow environments.

[0355] The following describes an exemplary unmanned aerial vehicle (UAV) control unit provided by an embodiment of the present invention. Figure 3 This is a schematic diagram of an exemplary hardware architecture of a drone control unit provided in an embodiment of the present invention.

[0356] In some embodiments, the UAV control unit may be an electronic device, or the UAV control unit may include an electronic device. The electronic device includes a processor, memory, and a network interface connected via a system bus. The processor of the electronic device provides computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the electronic device stores data. The network interface of the electronic device is used to communicate with other external terminals or servers via a network connection. In some embodiments, the network interface may be a wired network interface; in some embodiments, the network interface may also be a wireless network interface. When the computer program is executed by the processor, it implements the methods in the embodiments of the present invention.

[0357] Those skilled in the art will understand that Figure 3 The architecture shown is merely a block diagram of a portion of the architecture related to the present invention and does not constitute a limitation on the electronic device to which the present invention is applied. Specific electronic devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0358] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

[0359] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0360] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on an electronic device, all or part of the processes or functions described in the embodiments of the present invention are generated. The electronic device may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0361] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for structural flaw detection of unmanned aerial vehicles based on acoustic signature and spectrum, characterized in that, Applied to a drone control unit, the method includes: Control the drone to fly and hover at the target detection location of the structure to be tested; The controllable excitation device of the UAV outputs mechanical excitation pulses to the structure under test to excite the structure under test to generate forced radiated sound waves. The acoustic sensor array of the UAV is used to collect raw acoustic signals containing the forced radiated sound waves and the noise of the UAV platform. Obtain the current real-time rotor speed data of the UAV; The real-time rotor speed data is used as a dynamic environmental noise reference, and the original acoustic signal is subjected to noise cancellation processing to obtain the target pure acoustic signal. Acoustic spectral features are extracted from the target pure acoustic signal to obtain a target acoustic feature parameter set; The target acoustic feature parameter set is input into a pre-trained structural damage recognition model to obtain the damage determination result for the structure under test.

2. The method according to claim 1, characterized in that, The acoustic spectral feature extraction of the target pure acoustic signal yields a target acoustic feature parameter set, including: Perform a Fast Fourier Transform on the target pure acoustic signal to obtain a frequency domain signal; Extract the main resonant frequency parameter, high-frequency energy attenuation coefficient, and bandwidth energy ratio parameter from the frequency domain signal; The target acoustic feature parameter set is obtained by splicing and combining the main resonant frequency parameter, the high-frequency energy attenuation coefficient, and the frequency band energy ratio parameter.

3. The method according to claim 2, characterized in that, The extraction of the main resonant frequency parameter, high-frequency energy attenuation coefficient, and bandwidth energy ratio parameter from the frequency domain signal includes: Extract the first integral total energy value of the frequency domain signal within a first preset frequency range; Extract the second integral energy total value of the frequency domain signal within a second preset frequency interval, wherein the frequency band of the first preset frequency interval is higher than the frequency band of the second preset frequency interval. The ratio of the first total integrated energy value to the second total integrated energy value is calculated to obtain the frequency band energy ratio parameter.

4. The method according to claim 1, characterized in that, The controllable excitation device that controls the UAV outputs mechanical excitation pulses to the structure under test, including: The drone is controlled to approach the structure under test until the flexible buffer mechanism at the front end of the controllable excitation device abuts the surface of the structure under test. The system acquires real-time compression data of the flexible buffer mechanism, and in response to the real-time compression data reaching a preset buffer threshold, controls the UAV to output anti-recoil compensation thrust using its own flight control system to maintain the contact state of the flexible buffer mechanism and the current hovering attitude. While maintaining the current hovering posture, the electromagnetic catapult component in the controllable excitation device is controlled to output the mechanical excitation pulse to the structure under test through the flexible buffer mechanism.

5. The method according to claim 4, characterized in that, The controllable vibration device's electromagnetic ejection assembly outputs the mechanical vibration pulse to the structure under test via the flexible buffer mechanism, including: Obtain the target material type of the structure under test; If the target material type is the first material, then the electromagnetic catapult assembly is controlled to output a first mechanical excitation pulse to the structure under test with a first driving current and a first pulse width; If the target material type is the second material, then control the electromagnetic catapult assembly to output a second mechanical excitation pulse to the structure under test with a second driving current and a second pulse width; Wherein, the acoustic impedance of the first material is greater than that of the second material, the first driving current is greater than that of the second driving current, and the first pulse width is less than that of the second pulse width.

6. The method according to claim 1, characterized in that, The step of inputting the target acoustic feature parameter set into a pre-trained structural damage recognition model to obtain a damage determination result for the structure under test includes: The target acoustic feature parameter set is constructed into a one-dimensional feature vector and input into the structural damage identification model for feature mapping analysis; Obtain the anomaly confidence probability value for each candidate damage category output by the structural damage identification model; If the maximum anomaly confidence probability value is greater than the preset alarm threshold, then a damage determination result containing the internal damage type and the estimated damage size is generated.

7. The method according to any one of claims 1-6, characterized in that, The real-time rotor speed data includes the independent real-time speed data of each rotor unit when the UAV is currently in differential hovering state. The step of using the real-time rotor speed data as a dynamic environmental noise reference and performing noise cancellation processing on the original acoustic signal to obtain the target clean acoustic signal includes: Based on the real-time rotational speed data of each rotor unit and the number of blades configured in each rotor unit, the dynamic blade passing frequency and its harmonic frequencies of each rotor unit are calculated. The dynamic blades corresponding to each rotor unit are aggregated by frequency and their harmonic frequencies to generate a multi-band dynamic noise reference matrix that is strongly correlated with the current hovering attitude. The multi-band dynamic noise reference matrix is ​​input as a reference signal to a multi-reference source adaptive filter to track and cancel the aliased wandering aerodynamic noise generated by the differential operation of multiple independent rotors in the original acoustic signal, thereby obtaining the target pure acoustic signal.

8. An electronic device, characterized in that, Used as a control unit for unmanned aerial vehicles, it includes one or more processors and memory; The memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the electronic device to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium storing computer instructions, characterized in that, When the computer instructions are executed on an electronic device, the electronic device causes the electronic device to perform the method as described in any one of claims 1-7.

10. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed on an electronic device, the electronic device causes the electronic device to perform the method as described in any one of claims 1-7.

Images