A non-coplanar rotor layout noise reduction structure and method for a multi-rotor unmanned aerial vehicle

Abstract

The application discloses a multi-rotor unmanned aerial vehicle non-coplanar rotor layout noise reduction structure and noise reduction method, relates to the technical field of unmanned aerial vehicle aerodynamic noise control, and comprises a sound and vibration sensing array, which is used for collecting acoustic signals and vibration signals generated by tail flow interference between adjacent rotor units when each rotor unit operates; and a telescopic arm assembly, each rotor unit is connected to the fuselage through one telescopic arm assembly, and the telescopic arm assembly can change the axial length thereof in response to a control instruction. The multi-rotor unmanned aerial vehicle non-coplanar rotor layout noise reduction structure and noise reduction method establish a complete method from interference modal identification, spatial pose dynamic matching to effect closed-loop verification from the engineering practice level, change the blind trial and error mode of presetting geometric parameters and then simulating and verifying, and make the noise reduction design of the non-coplanar rotor layout have the technical features of repeatability, predictability and self-adaptive optimization.

Description

Technical Field

[0001] This invention relates to the field of aerodynamic noise control technology for unmanned aerial vehicles (UAVs), specifically to a noise reduction structure and method for a non-coplanar rotor layout of a multi-rotor UAV. Background Technology

[0002] Multi-rotor drones, with their advantages of vertical takeoff and landing, stable hovering, and high maneuverability, have been widely used in aerial photography, logistics delivery, agricultural plant protection, and urban air transportation. However, with the rapid increase in the density of drone operations and the frequency of close proximity to people, aerodynamic noise has become increasingly prominent, becoming a key technical bottleneck restricting their airworthiness certification and public acceptance. Among the many noise sources, aerodynamic noise generated by rotor rotation is dominant, especially when multiple rotors are operating simultaneously, the complex aerodynamic interference between adjacent rotors significantly amplifies the noise level.

[0003] Traditional multi-rotor UAVs generally employ a coplanar rotor configuration, meaning that the rotation planes of all rotors are parallel and at approximately the same altitude. In this configuration, the tip vortex and wake generated by any rotor inevitably interfere with the rotation planes of adjacent rotors, inducing periodic load pulsations and broadband eddy noise. When the spacing between adjacent rotors is relatively small, the wake interference effect is particularly strong, leading not only to a decrease in aerodynamic efficiency but also to a significantly higher overall sound pressure level than when a single rotor operates independently. Existing noise reduction technologies mainly focus on two approaches: one is optimizing the blade shape for a single rotor, such as using swept tips, serrated trailing edges, or low-noise airfoils to weaken the tip vortex intensity; the other is through active control strategies, such as adjusting the rotational speeds of each rotor to stagger their rotation frequencies, thereby avoiding coherent superposition of sound energy. However, these methods do not change the inherent coplanar interference geometry between rotors and cannot fundamentally suppress the direct impact and shearing effects of the wake in the horizontal direction.

[0004] Furthermore, existing non-coplanar rotor layout designs primarily focus on improving hovering efficiency or wind resistance stability, lacking a systematic understanding of noise reduction mechanisms. Specifically, the vertical spacing between the upper and lower rotors, the horizontal misalignment angle, the combination of rotational directions, and the speed matching relationship—how ​​exactly do these factors affect the spectral characteristics and radiation directivity of wake interference noise? There is a lack of clear theoretical models and systematic experimental data to support this. Currently, a design principle for non-coplanar rotor layouts that directly aims at noise suppression and comprehensively considers both aerodynamic and acoustic performance has not yet been established.

[0005] Therefore, it is urgent to propose a noise reduction structure and corresponding control method for non-coplanar rotor layouts oriented towards noise suppression, so as to break the coherent generation conditions of large-scale vortex structures from the aerodynamic source and achieve efficient and low-cost noise reduction effect. Summary of the Invention

[0006] The purpose of this invention is to provide a noise reduction structure and method for a non-coplanar rotor layout of a multi-rotor unmanned aerial vehicle (UAV) to solve the problems mentioned in the background art.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a noise reduction structure for a non-coplanar rotor layout of a multi-rotor unmanned aerial vehicle, comprising a fuselage and multiple rotor units mounted on the fuselage, and further comprising:

[0008] The acoustic and vibration sensing array is used to collect acoustic and vibration signals generated by wake interference between adjacent rotor units during operation.

[0009] A telescopic arm assembly, wherein each rotor unit is connected to the fuselage via a telescopic arm assembly, the telescopic arm assembly being able to change its axial length in response to control commands to adjust the spatial height position of the rotor unit relative to other rotor units;

[0010] The flight control computer has a built-in noise reduction control module. The noise reduction control module receives the signals collected by the acoustic and vibration sensing array, analyzes the physical mode of the current wake interference based on the signals, and then generates the control command based on the physical mode to drive the telescopic arm assembly to dynamically adjust the spatial height position of the corresponding rotor unit, so as to change the spatial relative attitude between adjacent rotor units.

[0011] Preferably, the noise reduction control module pre-stores a modal-adjustment strategy library. The noise reduction control module calls the corresponding adjustment strategy from the modal-adjustment strategy library based on the parsed physical mode. This adjustment strategy specifies the adjustment direction and adjustment step size of the telescopic arm assembly, as well as the reconfiguration rules for the rotor unit's rotational speed offset and rotational direction. Furthermore, this adjustment strategy specifies the following specific parameters:

[0012] Adjustment direction: Based on the modal type, instruct the telescopic boom to extend or shorten.

[0013] Adjustment step size: The step size is expressed as a percentage of the rotor blade length. The basic step size is 2% to 5% of the blade length, and the maximum cumulative displacement of a single adjustment does not exceed 15% of the blade length.

[0014] Speed ​​offset: The offset is ±3% to ±8% of the current speed, with 1% as the minimum adjustment level.

[0015] Rotation direction reconfiguration rule: When adjacent rotors generate crosstalk modes, the rotation direction of one of the rotors is reversed, so that the adjacent rotors are in a state of opposite rotation.

[0016] Preferably, the physical modes include vortex ring impact mode, propeller tip vortex crosstalk mode, and helical wake entanglement mode;

[0017] The quantization criteria for the above three physical modes are as follows:

[0018] Vortex ring impact mode: When the angle between the axis of the tip vortex core of the upper rotor and the plane of rotation of the lower rotor is greater than 70 degrees, and the broadband noise signal collected by the miniature microphone below the motor mount of the lower rotor shows a continuous spectrum energy rise in the range of 2 to 5 times the rotor passing frequency, and at the same time the vibration signal output by the piezoelectric thin film sensor shows irregular random impact pulses, it is determined to be the vortex ring impact mode.

[0019] Tip vortex crosstalk mode: When the closest horizontal distance between the tip vortices of two adjacent rotors is less than half the blade length, and the cross-correlation function between the two microphone signals shows a significant negative peak near zero time delay, while the autocorrelation envelope of the vibration signal exhibits periodic decaying oscillation, it is determined to be tip vortex crosstalk mode.

[0020] Helical wake entanglement mode: When the ratio of the circumferential velocity component to the radial velocity component generated by the downwash of the upper rotor at the plane of the lower rotor is greater than 2, and the power spectrum of the microphone signal shows harmonic peaks at fractional harmonics of the rotor passing frequency, such as 0.5 times or 1.5 times, and at the same time the time-domain waveform of the piezoelectric thin film sensor signal shows low-frequency (below the rotor passing frequency) modulated envelope fluctuations, it is determined to be a helical wake entanglement mode;

[0021] The aforementioned adjustment strategy employs different vortex interference suppression mechanisms for different physical modes:

[0022] When the vortex ring impact mode is identified, the spatial height position of the rotor unit is adjusted so that the rotation plane of the lower rotor actively cuts into the vortex core region of the tip vortex of the upper rotor, forcing the vortex core to be sheared and broken before impact.

[0023] When the propeller tip vortex crosstalk mode or the propeller wake entanglement mode is identified, the spatial height position is adjusted to make the wake of the upper rotor deviate from the plane of rotation of the lower rotor, thereby offsetting the axes of the two vortices and avoiding direct interference.

[0024] Preferably, after the initial adjustment, the noise reduction control module receives a new signal collected by the acoustic vibration sensing array again and calculates the interference attenuation ratio between the signal collected before adjustment and the new signal. When the interference attenuation ratio is lower than a preset qualified threshold, the noise reduction control module repeats the action of parsing and generating control commands until the interference attenuation ratio reaches or exceeds the qualified threshold.

[0025] Preferably, the acoustic vibration sensing array includes multiple miniature microphones and multiple piezoelectric thin film sensors;

[0026] The miniature microphones are placed on the lower surface of the motor mount of each rotor unit to collect broadband noise signals;

[0027] The piezoelectric thin-film sensor is attached to the fuselage skin between adjacent rotor units to sense structural vibration signals caused by wake interference.

[0028] Preferably, the telescopic boom assembly includes a fixed sleeve, a telescopic boom, and an electrostrictive actuator;

[0029] The fixed sleeve is installed on the fuselage, the telescopic boom is slidably sleeved inside the fixed sleeve, the rotor unit is installed on the outer end of the telescopic boom, and the electrostrictive actuator is disposed between the fixed sleeve and the telescopic boom for driving the telescopic boom to slide axially.

[0030] Preferably, the control command output by the flight control computer is a current or voltage signal, and the electrostrictive actuator generates a corresponding axial displacement according to the magnitude of the signal.

[0031] A noise reduction method for a non-coplanar rotor configuration of a multi-rotor UAV, applied to the noise reduction structure of the non-coplanar rotor configuration of the multi-rotor UAV, includes the following steps:

[0032] Step 1: Acquire the initial signals generated by wake interference during the operation of each rotor unit using an acoustic and vibration sensing array;

[0033] Step two: The noise reduction control module of the flight control computer receives the initial signal and analyzes the physical mode of the current wake interference.

[0034] Step 3: Generate control commands based on the physical modal to drive the telescopic arm assembly to change the spatial height position of the corresponding rotor unit, so as to adjust the spatial relative pose between adjacent rotor units.

[0035] Preferably, step three further includes:

[0036] Step four: Collect the adjusted new signal again through the acoustic vibration sensing array, and calculate the interference attenuation ratio between the initial signal and the new signal;

[0037] Step 5: Compare the interference attenuation ratio with a preset qualified threshold. If the interference attenuation ratio is lower than the qualified threshold, proceed to Step 2: re-analyze the physical mode based on the new signal and generate new control commands until the interference attenuation ratio reaches or exceeds the qualified threshold.

[0038] Preferably, the specific method for analyzing the physical modes in step two is as follows:

[0039] The initial signal is input into a pre-trained convolutional neural network model, which outputs the dominant mode label corresponding to the current wake interference. The dominant mode label is one of the following: vortex ring impact mode, blade tip vortex crosstalk mode, or helical wake entanglement mode.

[0040] This invention provides a noise reduction structure and method for a non-coplanar rotor layout of a multi-rotor unmanned aerial vehicle (UAV). It offers the following advantages:

[0041] The system collects acoustic and vibration signals generated by the wake interference between rotors in real time using an acoustic and vibration sensing array. It then uses a convolutional neural network to identify the current physical modes of interference online, such as vortex ring impact mode, tip vortex crosstalk mode, or spiral wake entanglement mode. Based on a mode-adjustment strategy library, it generates corresponding asymmetric adjustment commands to drive the retractable arm assembly to dynamically change the spatial altitude position of the rotors, while simultaneously fine-tuning the rotation speed and reconfiguring the rotation direction. This closed-loop dynamic adjustment mechanism ensures that the spatial relative attitude between the rotors always matches the wake field characteristics under the current flight conditions. Regardless of whether the UAV is hovering, in low-speed level flight, or in maneuvering flight, the system can actively cut the lower rotor's rotation plane into the vortex core region of the upper rotor's tip vortex or offset the two vortex axes in the vertical direction, thereby disrupting the coherent generation conditions of the large-scale vortex structure. This reduces the amplitude fluctuation range of interference noise in different flight speed ranges of the UAV compared to the control model with a static non-coplanar layout. Furthermore, when the flight conditions change abruptly, the system can reconverge to a low-noise state within a few control cycles without ground intervention or manual calibration, thus solving the defect of the static optimal layout where the noise reduction effect is drastically reduced in actual dynamic flight.

[0042] By combining a retractable arm assembly, an acoustic and vibration sensing array, and a noise reduction control module with an embedded modal-adjustment strategy library, a complete technical chain of perception-analysis-adjustment-verification is constructed. The modal-adjustment strategy library is built based on the mapping relationship between interference patterns and noise characteristics under various layout parameter combinations obtained from wind tunnel experiments, transforming the complex fluid-structure interaction problem into finite discrete modal categories and corresponding adjustment rules. The closed-loop iterative verification mechanism calculates the interference attenuation ratio before and after adjustment and compares it with a dynamic pass threshold, ensuring that each adjustment action brings quantifiable noise reduction benefits. When the attenuation ratio fails to meet the standard, the adjustment step size is automatically increased or the strategy is switched until the requirements are met. This not only directly serves the online noise reduction control of UAVs, but the accumulated adjustment parameters and modal discrimination thresholds can also serve as reference criteria in the ground design phase, providing data support for the subsequent static initial design of non-coplanar layouts. Therefore, from the perspective of engineering practice, a complete methodology has been established, from interference mode recognition to spatial pose dynamic matching, and then to effect closed-loop verification. This changes the blind trial-and-error mode of first setting geometric parameters and then verifying through simulation, and enables the noise reduction design of non-coplanar rotor layout to have the technical characteristics of repeatability, predictability, and adaptive optimization. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the overall structure of the noise reduction structure for the non-coplanar rotor layout of the multi-rotor UAV of the present invention;

[0044] Figure 2 This is a timing diagram of the control logic for the noise reduction method of the present invention.

[0045] In the diagram: 1. Fuselage; 2. Rotor unit; 3. Telescopic arm assembly; 31. Fixed sleeve; 32. Telescopic arm; 33. Electrostrictive actuator; 4. Acoustic and vibration sensing array; 41. Miniature microphone; 42. Piezoelectric thin film sensor; 5. Flight control computer. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Please see Figure 1 and Figure 2This invention provides a technical solution: a noise reduction structure for a non-coplanar rotor layout of a multi-rotor UAV, wherein multiple arm mounting seats are uniformly arranged circumferentially on the main body of the fuselage 1. Each arm mounting seat is equipped with a set of telescopic arm assemblies 3. The telescopic arm assembly 3 consists of a fixed sleeve 31, a telescopic arm 32, and an electrostrictive actuator 33. The root of the fixed sleeve 31 is rigidly connected to the arm mounting seat, and the telescopic arm 32 is coaxially slidably inserted into the fixed sleeve 31. A rotor motor mount is installed at the outer end of the telescopic arm 32, and the rotor unit 2 is fixed to the motor mount. The electrostrictive actuator 33 is a drive rod made of piezoelectric ceramic stack or super magnetostrictive material, one end of which abuts against the inner wall of the closed end of the fixed sleeve 31, and the other end is connected to the inner end of the telescopic arm 32. When a voltage or current control command is applied to the electrostrictive actuator 33, the drive rod extends or retracts axially, thereby pushing the telescopic arm 32 to slide relative to the fixed sleeve 31, changing the spatial height position of the rotor unit 2 relative to the fuselage 1 and other rotor units 2.

[0048] The acoustic vibration sensing array 4 includes two types of sensors. The first type is a miniature capacitive MEMS microphone, with three of these microphones installed at equal angular intervals along the circumferential direction on the lower surface of each rotor motor mount, with their sensing surfaces facing downwards. These microphones are used to collect broadband aerodynamic noise signals in the near-field region below the rotor, especially the beat frequency components and broadband turbulent energy generated when adjacent rotor wakes collide. The second type is a piezoelectric thin-film sensor 42, made of polyvinylidene fluoride film. This film is attached to the inner surface of the fuselage skin between two adjacent arm mounts. It is used to sense structural vibration signals transmitted to the fuselage shell by inter-rotor wake interference. The time-domain impact period and frequency-domain peak distribution of this vibration signal can reflect the passing frequency of the interference vortices.

[0049] The noise reduction control module built into the flight control computer 5 receives the raw time-domain signals from all microphones and piezoelectric thin-film sensors 42 in real time. The module first performs DC-free and bandpass filtering on the signals, with the filtered passband covering the rotor's passing frequency and its first and second harmonic frequencies. Subsequently, the module extracts a set of characteristic parameters from the filtered signals, including: the peak cross-correlation value and hysteresis time between the corresponding microphone signals from adjacent rotors, the slope of the attenuation envelope of the autocorrelation function of the piezoelectric thin-film sensor 42 signal, and the energy ratio in a specific narrowband frequency band. These characteristic parameters constitute the interference eigenvector at the current moment.

[0050] The noise reduction control module contains a built-in convolutional neural network model, which has been pre-trained using wind tunnel experimental data. During training, interference feature vectors are simultaneously collected under different combinations of rotor height differences, misalignment angles, and rotational speed differences, and the physical morphology of the wake interference is manually labeled. The labeled categories include "vortex ring impact mode," "tip vortex crosstalk mode," and "spiral wake entanglement mode." In actual flight, the module inputs the real-time extracted interference feature vectors into the network, and the network outputs a probability distribution. The category with the highest probability is selected as the dominant physical mode of the current wake interference.

[0051] Based on the identified dominant physical mode, the noise reduction control module searches for the corresponding adjustment scheme from a pre-stored mode-adjustment strategy library. This strategy library exists in the form of a condition-action rule table. For example, if the current mode is "vortex ring impact mode," the strategy requires extending the retractable arm assembly 3 corresponding to the lower rotor by one adjustment step, so that the rotation plane of the lower rotor moves closer to the vortex core region of the tip vortex of the upper rotor, while reducing the rotational speed of the lower rotor by a preset small offset. If the current mode is "tip vortex crosstalk mode," the strategy requires raising one of the two adjacent rotors and lowering the other, while adjusting the rotational phase difference between the two to opposite directions. If the current mode is "spiral wake entanglement mode," the strategy requires alternating and staggering the heights of all rotors and redistributing the rotational direction so that the rotational directions of adjacent rotors are opposite.

[0052] The adjustment command is sent to the drive circuit of the corresponding electrostrictive actuator 33 in the form of a pulse width modulation signal. The drive circuit adjusts the voltage amplitude applied to the piezoelectric ceramic according to the pulse width, thereby controlling the elongation of the drive rod. At the same time, the command also adjusts the speed of the corresponding rotor motor through the electronic speed controller. The entire adjustment process is completed in multiple consecutive control cycles. After each control cycle, the acoustic and vibration sensing array 4 re-acquires the signal, and the noise reduction control module extracts the interference feature vector again and compares it with the vector before adjustment. If the cross-correlation peak value in the interference feature vector decreases significantly and the autocorrelation envelope slope becomes gentler, the current adjustment amount is maintained; otherwise, the next round of adjustment is executed according to the strategy until the interference feature vector reaches a stable state. All the above calculations and control processes are completed in real time on the UAV's flight control computer 5 without ground intervention.

[0053] The noise reduction control module internally allocates a non-volatile storage area, within which a modality-adjustment strategy library is stored. This strategy library is constructed in the form of a multi-branch conditional mapping table, where each row corresponds to a predefined wake interference physical mode, and each column corresponds to a type of action command parameter for the adjustment actuator. The first column of the mapping table records the modality identifier code, used for matching with the modality label output by the convolutional neural network; the second column records the adjustment direction of the telescopic arm assembly 3, including binary commands for extension or retraction.

[0054] The third column records the adjustment step size, which is represented by the incremental levels of the drive voltage of the electrostrictive actuator 33. The axial displacement corresponding to each level should make the displacement of the telescopic boom 32 reach 2% to 5% of the blade length, i.e., a basic step size. This displacement is determined through experimental calibration.

[0055] The fourth column records the speed offset, which is a specific value as a percentage adjustment of the target motor speed relative to the current speed. However, this value is stored in the strategy library as discrete levels, such as small negative offset, zero offset, and small positive offset. The fifth column records the rotation direction reconfiguration rule, which is represented by a Boolean flag. When the flag is true, two phases in the three-phase wiring sequence of the corresponding motor are swapped, thereby changing the rotation direction of the rotor.

[0056] During actual flight, once the noise reduction control module obtains the dominant physical mode output by the convolutional neural network, the strategy caller within the module uses the mode's identifier as an index to perform a hash lookup in the mode-adjustment strategy library mapping table, directly locating the corresponding strategy line. Then, the module parses the adjustment direction, step size, speed offset, and direction reconfiguration flag in the strategy line into specific execution instructions: the adjustment direction and step size are converted into the duty cycle increment / decrement value and increment count of the pulse width modulation signal applied to the corresponding electrostrictive actuator 33 drive circuit; the speed offset is converted into a correction coefficient for the corresponding electronic speed controller's speed setpoint; and the direction reconfiguration flag is converted into a switching signal, triggering a commutation operation by the relay group connected to the motor's three-phase lines. All parsed instructions are packaged in the form of data frames and sent in parallel to the actuators' drivers via the flight control computer 5's general-purpose input / output interface.

[0057] To prevent mismatches between the preset conditions in the policy library and the actual flight environment, the noise reduction control module also includes a built-in policy update interface. In ground maintenance mode, the UAV's ground station software can add new modal-policy mapping entries or modify the step size and offset settings of existing entries in the policy library based on new wind tunnel test data or actual flight logs. During the update process, a wireless data transmission link is established between the ground station and the flight control computer 5 to upload and overwrite the encrypted policy table data block to the corresponding address in the non-volatile storage area. Simultaneously, the module retains a factory default policy library as a fallback option. If the uploaded data verification fails or an actuator response anomaly occurs, it automatically switches to the default policy library to continue operation, preventing the noise reduction function from completely failing due to policy library corruption.

[0058] The noise reduction control module classifies the wake interference physical modes output by the convolutional neural network into one of three preset types: vortex ring impact mode, tip vortex crosstalk mode, and helical wake entanglement mode. The vortex ring impact mode is characterized by the tip vortex generated by the upper rotor forming a near-perpendicular impact angle with the plane of rotation of the lower rotor during downstream transport, causing the vortex core to suddenly rupture upon contact with the lower rotor blades, inducing broadband impact noise. The tip vortex crosstalk mode is characterized by the tip vortices generated by adjacent rotors approaching each other horizontally and undergoing shearing action, forming a narrow low-pressure zone between the two vortices. The vorticity exchange occurs periodically within this zone, generating discrete beat frequency noise. The helical wake entanglement mode is characterized by the overall downward spiral motion of the upper rotor's downwash, which generates a continuous circumferential non-uniform load when it enters the rotation region of the lower rotor, causing low-frequency modulated noise radiation.

[0059] For the three modes mentioned above, different spatial pose adjustment schemes are stored in the adjustment strategy library. When the vortex ring impact mode is identified, the noise reduction control module calls the corresponding adjustment strategy: selecting the lower of the two adjacent rotors involved in the impact as the adjustment target, and driving its telescopic arm assembly 3 to perform an extension action, so that the rotation plane of the lower rotor moves upward by one adjustment step, until the rotation plane is at the same height as the vortex core axis of the tip vortex of the upper rotor. At this time, the blade tip of the lower rotor will directly cut through the vortex core region during rotation, forcing the vortex core to be sheared into multiple small-scale vortex tubes by the induced velocity field generated by the blade tip before it impacts the blade. The energy dissipation rate of these small-scale vortex tubes is much faster than that of the complete vortex core, thus preventing the formation of impact noise at the source. At the same time, this adjustment strategy will also temporarily reduce the rotation speed of the lower rotor by one level, so that the cut-in speed matches the circulation of the vortex core, avoiding secondary vortex shedding due to excessively fast cut-in.

[0060] When the crosstalk mode is identified as a tip vortex, the adjustment strategy invoked by the noise reduction control module differs from that of the vortex ring impact mode. This strategy simultaneously adjusts the two adjacent rotors generating the crosstalk: extending the telescopic boom 32 corresponding to one rotor while shortening the telescopic boom 32 corresponding to the other rotor, creating an alternating height difference between the two in the vertical direction. This alternating height difference forces the vortex core axes of the two tip vortices to no longer be coplanar in space, transforming the original direct shearing effect in the horizontal direction into an indirect induced effect with a certain vertical spacing. Due to the existence of the vertical spacing, the energy exchange efficiency between the two vortices is reduced, and the vorticity cannot be concentrated in the narrow gap, thereby suppressing the periodic amplitude fluctuations of the beat frequency noise. In addition, this adjustment strategy also sets the rotation directions of the two rotors to be opposite, causing the induced velocity directions of the two at the alternating height layer to cancel each other out, further weakening the crosstalk intensity.

[0061] When a spiral wake entanglement mode is identified, the noise reduction control module invokes a third adjustment strategy. This strategy involves gradually increasing the vertical distance between the upper and lower rotor blades through multiple adjustments, while simultaneously adjusting the rotational speed of the upper rotor one level higher and the rotational speed of the lower rotor one level lower. The increased vertical distance allows the spiral tube generated by the upper rotor a longer travel distance before reaching the plane of the lower rotor for diffusion and dissipation, during which the circumferential velocity component of the tube attenuates due to viscosity. The differentiated adjustment of the rotational speeds of the upper and lower rotors disrupts the fixed phase-locked relationship between the spiral tube and the lower rotor, making the timing of the tube's impact on the lower blade no longer strictly periodic. This redistributes the low-frequency modulated noise energy across a wider frequency range, reducing the sound pressure level peak in the sensitive frequency band. After the adjustment strategies for the three modes mentioned above are executed, the noise reduction control module will re-acquire acoustic and vibration signals for verification. If the identified mode type has not changed or still belongs to the same category, the next round of adjustment will continue according to the corresponding strategy until the modal characteristic parameters drop to the preset stable range.

[0062] After completing the initial adjustment, the noise reduction control module does not immediately terminate the noise reduction process. Instead, it initiates a closed-loop iterative verification subroutine. This subroutine first acquires the new signal within a complete acquisition window after adjustment from the acoustic and vibration sensing array 4, and simultaneously retrieves the initial signal stored before adjustment began. The module has a built-in interference attenuation ratio calculation unit. This unit estimates the power spectrum of both the initial and new signals, and defines a frequency range on the power spectrum where the wake interference is most concentrated. The lower limit of this range is taken as the rotor passing frequency, and the upper limit is taken as several octaves of the passing frequency.

[0063] The calculation unit calculates the total energy values ​​of the initial signal and the new signal within the frequency range, respectively. The ratio of the total energy value of the new signal to the total energy value of the initial signal is then used as the residual energy ratio. This residual energy ratio is subtracted from the constant 1 to obtain the frequency domain energy attenuation contribution. Simultaneously, the calculation unit extracts the time-domain envelope of the piezoelectric thin-film sensor 42's output signal, calculates the difference between the peak values ​​of the initial envelope and the new envelope, and divides this difference by the peak value of the initial envelope to obtain the time-domain vibration attenuation contribution.

[0064] The interference attenuation ratio is calculated as follows: the frequency domain energy attenuation contribution and the time domain vibration attenuation contribution are weighted according to the flight mode and then summed. Specifically, when the UAV is in hovering mode, the weight of the time domain vibration attenuation contribution is approximately 70%, and the weight of the frequency domain energy attenuation contribution is approximately 30%; when the UAV is in forward flight mode, the weight of the frequency domain energy attenuation contribution is approximately 70%, and the weight of the time domain vibration attenuation contribution is approximately 30%. The weighted sum of the two contributions is the interference attenuation ratio. The weight values ​​are pre-stored in the flight control computer 5 and can be calibrated according to actual tests.

[0065] The module also stores a pass / fail threshold register. The threshold in this register is not a fixed value, but is calculated and generated in real time by the flight control computer 5 based on the current flight status. The flight status includes attitude angles, vertical velocity, horizontal velocity, and the estimated mass of the onboard mission equipment.

[0066] The threshold calculation logic is as follows:

[0067] When the drone is hovering or flying at low speed and the load is symmetrically distributed, the threshold is set to a default medium level.

[0068] When the drone is flying forward or sideways at a large angle, due to the complex rotor inflow conditions and frequent changes in interference modes, the threshold is automatically lowered by one level to avoid the excessively stringent convergence requirements causing the adjustment system to oscillate continuously.

[0069] When the drone hovers near the ground and there is a reflective surface below, the threshold automatically increases by one level. This takes full advantage of the fact that rotor interference noise is more easily amplified by ground reflection at this time, forcing the system to find better adjustment parameters. Specifically, the qualified threshold is preset to 0.6 in hovering mode, 0.5 in forward flight mode, and 0.7 in near-ground hovering mode.

[0070] The noise reduction control module compares the calculated current interference attenuation ratio with the value in the qualified threshold register. If the interference attenuation ratio reaches or exceeds the threshold, the current adjustment result is deemed satisfactory, and the module stops further iteration. It locks the length parameters of the current telescopic arm assembly 3 and the rotational speed parameters of each rotor, while continuously monitoring the signal of the acoustic vibration sensing array 4. The noise reduction process is only retried when a new abrupt change occurs in the signal. If the interference attenuation ratio is below the threshold, the current adjustment effect is deemed insufficient. The module automatically uses the current new signal as the reference signal, re-executes the convolutional neural network for modal recognition, and generates a new round of control commands from the modal-adjustment strategy library based on the recognition results. The step size of the new round of adjustment can be dynamically adjusted based on the difference between the attenuation ratio and the threshold after the previous adjustment: the larger the difference, the larger the adjustment step size; the smaller the difference, the smaller the adjustment step size. After each round of adjustment, the interference attenuation ratio is recalculated and compared with the threshold. This process is repeated iteratively until the exit condition is met or the preset maximum number of iterations is reached. The maximum number of iterations can be preset to 5. If the threshold is not met after reaching the maximum number of iterations, the module will retain the adjustment parameter corresponding to the one with the highest interference attenuation ratio in each iteration as the optimal solution under the current operating condition, and send a prompt message to the ground station through the data transmission link of the flight control computer 5 to inform the maintenance personnel that the sensors or actuators need to be checked for faults.

[0071] The acoustic and vibration sensing array 4 consists of multiple miniature microphones 41 and multiple piezoelectric thin-film sensors 42. The motor mount corresponding to each rotor unit 2 is made of aluminum alloy, with an annular groove machined into its lower surface. Three through-holes are equally spaced along the circumference of the groove, and a miniature condenser microphone is installed in each through-hole. The microphone's sensing surface is flush with the lower surface of the motor mount, facing directly downwards, and is used to collect aerodynamic noise generated during rotor rotation. The microphone's signal lines are led out through wiring channels inside the motor mount, laid along the outer wall of the retractable arm assembly 3, and finally converged to the analog-to-digital conversion interface of the flight control computer 5. To reduce wind noise interference, each microphone also has a wind shield made of porous sintered metal at its front end, which is embedded in the through-hole entrance and remains flush with the lower surface of the motor mount.

[0072] The piezoelectric thin-film sensor 42 uses a polyvinylidene fluoride piezoelectric thin film as the sensing element. The film thickness is on the order of micrometers, and silver electrodes are deposited on both sides. One of these piezoelectric thin-film sensors 42 is adhered to the inner surface of the fuselage skin area between every two adjacent arm mounts. Before adhesion, a layer of flexible epoxy adhesive is applied to the inner surface of the fuselage skin. After the thin-film sensor is smoothly adhered, uniform pressure is applied. After the adhesive layer cures, the thin-film sensor forms a tight mechanical coupling with the fuselage skin. The spacing between two adjacent thin-film sensors is determined according to the dimensions of the fuselage 1, ensuring that at least one sensor covers the fuselage 1 area corresponding to each pair of adjacent rotors. The two electrode leads of the thin-film sensor are connected to a shielded cable via conductive silver adhesive. The cable is arranged along the internal reinforcing ribs of the fuselage 1 and connected to the input terminal of the charge amplification circuit of the flight control computer 5. This circuit converts the charge signal generated by the piezoelectric thin film into a voltage signal for analog-to-digital conversion sampling.

[0073] To distinguish noise signals generated by different rotor units 2, each microphone records its corresponding rotor number during the factory calibration phase, and a channel-to-number mapping table is established in the memory of the flight control computer 5. Similarly, the installation position of each piezoelectric thin-film sensor 42 also corresponds to a pair of adjacent rotor numbers. When the acoustic and vibration sensing array 4 is working, the signals from all microphones are sampled sequentially in a time-division multiplexing manner, and the signals from all piezoelectric thin-film sensors 42 are sampled synchronously in parallel. The sampled data is temporarily stored in the memory of the flight control computer 5 in the form of data frames, and the header of each data frame contains a timestamp and flight attitude information. When the noise reduction control module analyzes the signals, it associates the signals of each channel with specific rotors or rotor pairs according to the mapping table, thereby determining the spatial source of noise and vibration and providing spatial distribution information for subsequent interference mode identification. The above-mentioned microphone and thin-film sensor installation structure does not need to be disassembled throughout the entire life cycle of the UAV; the channel mapping relationship only needs to be recalibrated when the fuselage 1 or motor mount is replaced.

[0074] The fixed sleeve 31 of the telescopic boom assembly 3 is made of carbon fiber composite material through winding molding. Its inner wall is precision ground and coated with a layer of polytetrafluoroethylene (PTFE) anti-friction coating. A flange is provided at the root of the fixed sleeve 31, which is fixedly connected to the aluminum alloy mounting base on the fuselage 1 by bolts. A rubber shock-absorbing washer is sandwiched between the flange and the mounting base. The telescopic boom 32 is made of high-strength hollow aluminum alloy tubing, with a sliding clearance between its outer diameter and the inner diameter of the fixed sleeve 31. The outer surface of the telescopic boom 32 is also coated with PTFE, forming a mating friction pair with the coating on the inner wall of the fixed sleeve 31. One end of the telescopic boom 32 is located inside the fixed sleeve 31, and this end is machined with an annular groove for connecting to the output end of the electrostrictive actuator 33. The other end of the telescopic boom 32 extends outside the fixed sleeve 31, and a threaded hole is opened on its end face for mounting the rotor motor mount.

[0075] The electrostrictive actuator 33 employs a multi-layered stack of piezoelectric ceramics, with each ceramic piece having a uniform thickness. The total stack height matches the pre-reserved travel space inside the fixed sleeve 31. One end of the piezoelectric ceramic stack abuts against the inner wall of the closed end of the fixed sleeve 31 via an insulating gasket, while the other end connects to the annular groove of the telescopic arm 32 via a spherical connector. This spherical connector allows the telescopic arm 32 to automatically self-align under lateral forces, preventing the piezoelectric ceramic stack from being subjected to bending moments. The piezoelectric ceramic stack is surrounded by a flexible insulating potting compound. After curing, this potting compound maintains elasticity and fixes the relative position of the piezoelectric ceramic stack, while preventing relative misalignment between the ceramic pieces due to vibration and impact. Positive and negative electrode cables leading from the piezoelectric ceramic stack pass through sealed holes in the side wall of the fixed sleeve 31 and are connected to the drive circuit board of the flight control computer 5 via shielded wires.

[0076] The drive circuit board outputs an adjustable DC voltage. When this voltage is applied to both ends of the piezoelectric ceramic stack, the piezoelectric ceramics undergo expansion and contraction along the polarization direction. When the voltage increases, the piezoelectric ceramic stack elongates, pushing the telescopic arm 32 outward; when the voltage decreases, the piezoelectric ceramic stack contracts, and the telescopic arm 32 retracts inward under the combined action of gravity and aerodynamic reaction force of the rotor unit 2. To prevent the telescopic arm 32 from rigidly impacting the closed end of the fixed sleeve 31 during retraction, a layer of polyurethane buffer pad is attached to the inner wall of the closed end of the fixed sleeve 31, and the same buffer pad is also attached to the annular groove end face of the telescopic arm 32. The exposed parts of the entire telescopic arm assembly 3 are wrapped with heat-shrink tubing to prevent dust and moisture from entering the sliding gap. After assembly, a small amount of perfluoropolyether grease is injected into the fixed sleeve 31 to further reduce sliding friction and improve smoothness of movement.

[0077] After the noise reduction control module of the flight control computer 5 completes strategy generation, it converts the adjustment commands into analog voltage signals for output. Each retractable arm assembly 3 corresponds to an independent digital-to-analog converter channel. The output voltage range of this channel is from a preset minimum voltage to a maximum voltage, corresponding to the minimum and maximum extension of the electrostrictive actuator 33, respectively. The module maintains a voltage-displacement mapping table, which is obtained through a one-time calibration before leaving the factory: on the test fixture, multiple discrete voltage values ​​are applied to the electrostrictive actuator 33, and the end face displacement of the retractable arm 32 is measured using a laser displacement sensor. The correspondence between voltage and displacement is recorded as an interpolation curve. In actual flight, the module obtains the target voltage value by looking up the mapping table and interpolating backwards according to the axial displacement required by the strategy. Then, it writes the voltage value into the input register of the digital-to-analog converter in digital form.

[0078] The analog voltage signal output from the digital-to-analog converter (DAC) first undergoes impedance transformation via a voltage follower, and then is fed into a voltage amplification circuit composed of a high-voltage operational amplifier. Since the piezoelectric ceramic stack requires a relatively high driving voltage to generate sufficient elongation, the amplification circuit amplifies the DAC's output voltage (typically within the standard voltage range) to a preset high-voltage range. The amplified voltage signal is transmitted to the electrostrictive actuator 33 via a twisted-pair shielded cable. The shielding layer is grounded at the actuator end via a capacitor to filter out common-mode interference. A high-resistance bleeder resistor is connected in parallel across the piezoelectric ceramic stack to release residual charge on the piezoelectric ceramic after the flight controller is powered off, preventing fatigue caused by the telescopic boom 32 being in a continuously energized elongated state.

[0079] To accommodate varying response speed requirements, the noise reduction control module also supports outputting current signals as an alternative to control commands. When a current signal is selected, a voltage-to-current conversion circuit is connected after the digital-to-analog converter, employing a classic structure of operational amplifiers and transistor current amplification to output a standard industrial current signal range. Current signals exhibit superior attenuation resistance over long distances compared to voltage signals, making them particularly suitable for UAVs with longer arm lengths. On the electrostrictive actuator 33 side, the current signal is first converted to voltage via a precision sampling resistor, and then fed into the piezoelectric ceramic drive circuit via a voltage follower. Regardless of whether voltage or current mode is used, the drive circuit is designed with overvoltage and overcurrent protection functions. When the output voltage exceeds the rated voltage of the piezoelectric ceramic stack or the output current exceeds the safety threshold, the protection circuit automatically cuts off the output and feeds back the fault signal to the flight control computer 5.

[0080] While sending control commands, the flight control computer 5 also acquires the actual voltage or current values ​​fed back by the electrostrictive actuator 33 in real time through an independent analog-to-digital conversion channel. This feedback value is compared with the target value by a proportional-integral controller. When the deviation between the two exceeds the allowable range, the module automatically adjusts the output value of the digital-to-analog converter to compensate for errors caused by line voltage drop or temperature drift. This closed-loop command verification mechanism ensures that the electrical signal applied to the piezoelectric ceramic stack is always consistent with the value expected by the noise reduction control module, thereby making the actual axial displacement of the telescopic boom 32 precisely correspond to the target displacement required by the adjustment strategy.

[0081] A noise reduction method for a non-coplanar rotor configuration of a multi-rotor UAV, wherein the noise reduction method is implemented as follows:

[0082] First, step one is executed: After the UAV is powered on and completes its self-test, the flight control computer 5 starts the data acquisition thread of the acoustic and vibration sensing array 4. This thread simultaneously performs analog-to-digital conversion on the signals from all the miniature microphones 41 and piezoelectric film sensors 42 at a fixed sampling rate. The data acquired in each sampling cycle is packaged into a data block and stored in a circular buffer in memory. The duration of the acquisition window is set according to the rotor rotation cycle to ensure that it includes at least several complete rotation cycles, thereby capturing the periodic characteristics of wake interference. The acquisition of the initial signal is triggered after the UAV enters a stable flight state (e.g., hovering or uniform forward flight). The triggering condition is provided by the attitude calculation module of the flight control computer 5: when the rate of change of the UAV's attitude angle, rate of change of altitude, and rate of change of velocity are all lower than their respective threshold values ​​and the duration exceeds the preset stabilization time, it is determined to be a stable flight state, and the acquisition of the initial signal begins. After acquisition, the initial signal is stored in a dedicated read-only buffer as a reference for subsequent interference attenuation calculations.

[0083] Step two follows immediately after step one. The noise reduction control module reads the most recent initial signal data block from the circular buffer and performs preprocessing: removing the DC component, applying a Hanning window to reduce spectral leakage, and then converting the time-domain signal to the frequency domain using a Fast Fourier Transform. Simultaneously, the module extracts the time-domain envelope from the signal from the piezoelectric thin-film sensor 42 and calculates the envelope using a Hilbert transform. The preprocessed multi-channel signal is input to a pre-trained convolutional neural network model. This model is compressed before deployment, and its weight parameters are stored in the flash memory of the flight control computer 5 in fixed-point form, ensuring that inference on the embedded processor is completed within one control cycle. The model's input layer receives a multi-channel feature map, which is composed of a power spectrum segment from each microphone channel and an envelope segment from each piezoelectric thin-film channel. Specifically, the feature vector is formed by taking the rotor passage frequency and 10 frequency points within the range of 2 to 5 harmonics in the power spectrum of each microphone channel, and 20 time-point samples of the envelope of each piezoelectric thin-film channel within one rotation cycle.

[0084] After the model's forward propagation calculation, the output layer generates a three-dimensional vector, with each dimension's value ranging from zero to one, corresponding to the probabilities of the vortex ring impact mode, the tip vortex crosstalk mode, and the spiral wake entanglement mode, respectively. The module selects the mode corresponding to the highest probability as the physical mode of the current wake interference. If the maximum probability value is lower than the preset confidence threshold, it is determined to be a mode with no significant interference. In this case, no further adjustments are made, the noise reduction process is temporarily suspended, and step one is restarted in the next acquisition window. The preset confidence threshold is 0.6.

[0085] Step 3 is executed after the physical mode is successfully identified in Step 2. The noise reduction control module searches for the corresponding control instruction template from the mode-adjustment strategy library in memory based on the identified mode type. For each mode, the template predefines the target object to be adjusted (e.g., the lower one of a pair of adjacent rotors causing interference, or adjusting both simultaneously), the target action to be adjusted (e.g., extending or shortening the telescopic arm assembly 3), and additional actions (e.g., adjusting the speed offset or changing the rotation direction). The module translates the actions in the template into specific actuator instructions: for the telescopic arm assembly 3, based on the target extension direction and step size, it queries the voltage-displacement mapping table to obtain the target value of the drive voltage and generates the corresponding digital output to the digital-to-analog converter; for speed adjustment, it calculates a new speed setpoint based on the offset size and sends it to the electronic speed controller of the corresponding rotor via a pulse width modulation signal; for rotation direction reconfiguration, it controls the operation of the motor commutation relay via a switching signal. All instructions are issued in parallel within the same control cycle to ensure that the spatial height position and speed of each rotor unit 2 are adjusted synchronously. During the adjustment process, the noise reduction control module continuously monitors the signal of the acoustic vibration sensing array 4 to prevent the adjustment action from triggering new interference modes. After the adjustment action is completed and the rotor speed stabilizes, the entire noise reduction method completes a full execution cycle, and then returns to the initial signal acquisition at the beginning of the step to prepare for possible closed-loop iteration or re-optimization after changes in operating conditions.

[0086] During the noise reduction process, after the command is issued in step three and both the telescopic arm assembly 3 and the rotor speed reach a stable state, the noise reduction method automatically proceeds to step four. Before step four begins, the noise reduction control module first waits for a preset stabilization delay. This delay length is determined by the response time of the electrostrictive actuator 33 and the adjustment time of the rotor motor speed controller. Typically, the stabilization delay is preset to between 0.5 and 1 second, and the specific value can be determined through ground calibration based on the actual response speed of the actuators, ensuring that all actuators have reached the target state and the flow field has been re-established and stabilized. After the stabilization delay ends, the module triggers the acoustic vibration sensing array 4 to perform a second data acquisition. The acquisition duration, sampling rate, and filtering parameters are completely consistent with the initial signal acquisition in step one to ensure the comparability between the two signals. The acquired new signal is stored in another read-only buffer, alongside the initial signal buffer.

[0087] Subsequently, the interference attenuation ratio calculation unit in the noise reduction control module simultaneously reads data from the initial signal buffer and the new signal buffer. The calculation process is divided into two parallel paths: frequency domain and time domain. In the frequency domain path, the unit performs a fast Fourier transform on both sets of signals to obtain power spectral density estimates. In the power spectrum, the unit automatically identifies a characteristic frequency band. The lower limit of this band is determined by the current rotor's passing frequency, while the upper limit is determined by finding the frequency in the power spectrum where the amplitude attenuates to the background noise level. The unit calculates the total energy of the initial signal and the new signal within this characteristic frequency band, and then records the ratio of the total energy of the new signal to the total energy of the initial signal as the energy residual ratio. A fixed value is then subtracted from this energy residual ratio to obtain the frequency domain attenuation contribution value. In the time domain path, the unit extracts the envelope of the signal from the piezoelectric thin film sensor 42, calculates the difference between the peak value of the initial envelope and the peak value of the new envelope, and then divides this difference by the peak value of the initial envelope to obtain a normalized peak attenuation rate. This attenuation rate is directly used as the time domain attenuation contribution value. Finally, the unit multiplies the frequency domain attenuation contribution value by a frequency domain weighting coefficient and the time domain attenuation contribution value by a time domain weighting coefficient, and adds the two together to obtain the final interference attenuation ratio. The sum of the frequency domain weighting coefficient and the time domain weighting coefficient is a fixed value, and their specific allocation is dynamically adjusted according to the current flight mode of the UAV: ​​in hovering mode, the time domain weighting coefficient is allocated higher, and in forward flight mode, the frequency domain weighting coefficient is allocated higher, in order to adapt to the changes in the dominance of noise and vibration under different flight attitudes.

[0088] Step five follows immediately after the interference attenuation ratio calculation. The noise reduction control module reads the threshold corresponding to the current operating condition from the qualified threshold register. This threshold remains unchanged during the execution of step four and will not be updated even if the flight status fluctuates slightly, to ensure the consistency of the judgment benchmark. The module compares the interference attenuation ratio calculated in step four with the threshold. If the interference attenuation ratio is equal to or greater than the threshold, the adjustment is considered successful. The noise reduction control module writes the current retractable arm assembly 3 length parameters, rotor speed parameters, and rotation direction parameters into the optimal parameter table in the non-volatile memory. At the same time, it clears the old data in the initial signal buffer and transfers the data in the new signal buffer as a new initial signal, which serves as the benchmark for the next stable flight state. After this, the noise reduction process enters a low-power monitoring mode, monitoring the average energy level of the acoustic vibration sensing array 4 at a low sampling rate. When the average energy level changes abruptly, the complete noise reduction process is restarted.

[0089] If the interference attenuation ratio is less than the threshold, it is determined that the current adjustment has not yet achieved the target effect. At this point, the noise reduction control module does not clear any buffers, but directly copies the contents of the new signal buffer to the initial signal buffer, overwriting the original initial signal. Then, the module automatically jumps back to step two and re-executes mode recognition based on the updated initial signal. When re-executing step three, the module calls an iterative step size adjustment logic: if this is the first iteration, the adjustment step size remains unchanged; if one iteration has already been performed and the attenuation ratio has increased compared to the previous round but still does not meet the target, the adjustment step size is reduced by one level to avoid overshoot; if the attenuation ratio has not increased compared to the previous round or has even decreased, the adjustment step size is increased by one level to accelerate convergence. The iterative step size adjustment logic ensures that the system can gradually approach the optimal adjustment parameters in multiple loops. The module also includes a maximum iteration counter. Each time the system jumps back to step two, the counter increments. When the counter reaches a preset upper limit, the iteration loop is forcibly terminated, and the adjustment parameter corresponding to the highest interference attenuation ratio in all iterations is retained as the suboptimal solution for the current operating condition. Simultaneously, a status alert is sent to the ground station. Before reaching the maximum iteration count, step five repeatedly drives the loop execution of steps two through four until the interference attenuation ratio meets the threshold requirement. The entire closed-loop iteration process is completed automatically within the flight control computer 5, without relying on intervention from ground operators.

[0090] In implementing the physical mode analysis in step two, a pre-trained convolutional neural network model is deployed within the noise reduction control module. The input data for this model comes from the initial signal acquired in step one. First, the module performs frame processing on the time-domain signal of each microphone channel, with each frame covering multiple rotor rotation cycles, and retaining partial overlap between adjacent frames to ensure temporal continuity. After applying a Hanning window to each frame signal, it is converted into a frequency domain power spectrum through a fast Fourier transform. For the signal from the piezoelectric film channel, the time-domain waveform within the same time window as the microphone signal is directly extracted, detrended, and normalized before being used as another input. The model adopts a one-dimensional convolutional neural network structure, with small-sized convolutional kernels and moderate strides in its convolutional layers, aiming to extract local peak patterns in the power spectrum and impact features in the time-domain waveform. After the convolutional layers, a global average pooling layer is connected instead of a fully connected layer to reduce the number of parameters, followed by two fully connected layers and a softmax output layer.

[0091] The model training process was completed on the ground. Training data was collected using the same sensor configuration as the UAV, conducted in an acoustic wind tunnel. A multi-rotor UAV platform was installed in the wind tunnel, capable of altering the height difference, offset angle, and rotational speed difference of each rotor via external actuators. Under different combinations of layout parameters, the sensor output signals and the wake interference physical morphology calibrated by a far-field microphone array and a high-speed particle image velocimetry system were recorded simultaneously. Each set of signal samples was manually labeled with its corresponding mode based on the calibration results: when an upper tip vortex impacted the lower rotor plane accompanied by broadband impact noise, it was labeled as a vortex ring impact mode; when two adjacent tip vortices were observed to be horizontally intertwined and generating discrete beat frequencies, it was labeled as a tip vortex crosstalk mode; when the downwash exhibited a helical tube shape and the lower rotor load fluctuated periodically, it was labeled as a helical wake entanglement mode. All samples, after time alignment and amplitude normalization, were proportionally divided into training, validation, and test sets.

[0092] The model training employs a cross-entropy loss function and an adaptive moment estimation optimizer, iterating multiple times on the training set until the loss on the validation set no longer decreases. After training, the model's weight and bias parameters are converted from floating-point to fixed-point numbers, and redundant connections with gradients close to zero are removed from the network, resulting in a lightweight model file. This file is burned into the flash memory of the flight control computer 5, along with a normalized parameter table used to map the real-time acquired signal amplitudes to the normalized range used during training.

[0093] In actual flight, when the model is invoked for inference in step two, the module first retrieves the initial signal for a complete time window from the circular buffer and performs frame segmentation, windowing, Fourier transform, and normalization according to the preprocessing procedure during training. The processed data is then fed into the model's input layer, where the model performs forward computation layer by layer, ultimately generating three node values ​​at the Softmax output layer. Each node value is between zero and one, and the sum of the three values ​​is one. The module takes the category corresponding to the maximum value as the dominant mode label for the current wake interference. If the maximum node value is lower than a preset confidence threshold, it indicates insufficient confidence in the model's judgment of the current mode. In this case, the module does not output any mode label but marks step two as invalid, and the noise reduction process returns to step one and re-acquires the initial signal. Only when the maximum node value reaches or exceeds the confidence threshold is the corresponding mode label output for subsequent steps. The entire inference process is completed within one control cycle, without affecting the real-time flight stability of the UAV.

[0094] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0095] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A noise reduction structure for a non-coplanar rotor layout of a multi-rotor unmanned aerial vehicle (UAV), comprising a fuselage and multiple rotor units mounted on the fuselage, characterized in that, Also includes: The acoustic and vibration sensing array is used to collect acoustic and vibration signals generated by wake interference between adjacent rotor units during operation. A telescopic arm assembly, wherein each rotor unit is connected to the fuselage via a telescopic arm assembly, the telescopic arm assembly being able to change its axial length in response to control commands to adjust the spatial height position of the rotor unit relative to other rotor units; The flight control computer has a built-in noise reduction control module. The noise reduction control module receives the signals collected by the acoustic and vibration sensing array, analyzes the physical mode of the current wake interference based on the signals, and then generates the control command based on the physical mode to drive the telescopic arm assembly to dynamically adjust the spatial height position of the corresponding rotor unit, so as to change the spatial relative attitude between adjacent rotor units.

2. The noise reduction structure for a non-coplanar rotor layout of a multi-rotor UAV according to claim 1, characterized in that, The noise reduction control module has a pre-stored modal-adjustment strategy library. The noise reduction control module calls the corresponding adjustment strategy from the modal-adjustment strategy library according to the parsed physical mode. The adjustment strategy specifies the adjustment direction and adjustment step size of the telescopic arm assembly, as well as the speed offset and rotation direction rematch rules of the rotor unit.

3. The noise reduction structure for a non-coplanar rotor layout of a multi-rotor UAV according to claim 2, characterized in that, The physical modes include vortex ring impact mode, propeller tip vortex crosstalk mode, and helical wake entanglement mode; The adjustment strategy targets different physical modes by adjusting the spatial height position of the rotor unit, so that the rotation plane of the lower rotor cuts into the vortex core region of the tip vortex of the upper rotor, or makes the wake of the upper rotor deviate from the rotation plane of the lower rotor.

4. The noise reduction structure for a non-coplanar rotor layout of a multi-rotor UAV according to claim 1, characterized in that, After the initial adjustment, the noise reduction control module receives the new signal collected by the acoustic vibration sensing array again and calculates the interference attenuation ratio between the initial signal and the new signal. When the interference attenuation ratio is lower than a preset qualified threshold, the noise reduction control module repeats the parsing and generating control command action until the interference attenuation ratio reaches or exceeds the qualified threshold.

5. The noise reduction structure for a non-coplanar rotor layout of a multi-rotor UAV according to claim 1, characterized in that, The acoustic vibration sensing array includes multiple miniature microphones and multiple piezoelectric thin film sensors; The miniature microphones are placed on the lower surface of the motor mount of each rotor unit to collect broadband noise signals; The piezoelectric thin-film sensor is attached to the fuselage skin between adjacent rotor units to sense structural vibration signals caused by wake interference.

6. The noise reduction structure for a non-coplanar rotor layout of a multi-rotor UAV according to claim 1, characterized in that, The telescopic boom assembly includes a fixed sleeve, a telescopic boom, and an electrostrictive actuator. The fixed sleeve is installed on the fuselage, the telescopic boom is slidably sleeved inside the fixed sleeve, the rotor unit is installed on the outer end of the telescopic boom, and the electrostrictive actuator is disposed between the fixed sleeve and the telescopic boom for driving the telescopic boom to slide axially.

7. A noise reduction structure for a non-coplanar rotor layout of a multi-rotor UAV according to claim 6, characterized in that, The control commands output by the flight control computer are current or voltage signals, and the electrostrictive actuator generates a corresponding axial displacement based on the magnitude of the signal.

8. A noise reduction method for a non-coplanar rotor layout of a multi-rotor unmanned aerial vehicle, applied to the structure described in any one of claims 1 to 7, characterized in that, Includes the following steps: Step 1: Acquire the initial signals generated by wake interference during the operation of each rotor unit using an acoustic and vibration sensing array; Step two: The noise reduction control module of the flight control computer receives the initial signal and analyzes the physical mode of the current wake interference. Step 3: Generate control commands based on the physical modal to drive the telescopic arm assembly to change the spatial height position of the corresponding rotor unit, so as to adjust the spatial relative pose between adjacent rotor units.

9. A noise reduction method for a non-coplanar rotor layout of a multi-rotor UAV according to claim 8, characterized in that, Step three is followed by: Step four: Collect the adjusted new signal again through the acoustic vibration sensing array, and calculate the interference attenuation ratio between the initial signal and the new signal; Step 5: Compare the interference attenuation ratio with a preset qualified threshold. If the interference attenuation ratio is lower than the qualified threshold, proceed to Step 2: re-analyze the physical mode based on the new signal and generate new control commands until the interference attenuation ratio reaches or exceeds the qualified threshold.

10. A noise reduction method for a non-coplanar rotor layout of a multi-rotor UAV according to claim 8, characterized in that, The specific method for analyzing physical modes in step two is as follows: The initial signal is input into a pre-trained convolutional neural network model, which outputs the dominant mode label corresponding to the current wake interference. The dominant mode label is one of the following: vortex ring impact mode, blade tip vortex crosstalk mode, or helical wake entanglement mode.

Images