A detection method and device for unmanned aerial vehicle production
By using multi-dimensional data synchronous acquisition and multi-modal fusion models, the problems of dynamic visual capture and acoustic detection anti-interference in the power system inspection of UAV production lines were solved, achieving high-precision and rapid fault identification and consistency detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WENZHOU UNIV OUJIANG COLLEGE
- Filing Date
- 2026-05-06
- Publication Date
- 2026-06-02
AI Technical Summary
Existing power system testing technologies on drone production lines suffer from problems such as difficulty in dynamic visual capture, poor acoustic detection anti-interference capabilities, and a lack of multi-dimensional data correlation judgment, resulting in poor testing results.
By employing multi-dimensional data synchronous acquisition technology, combined with stroboscopic light source illumination and acoustic signal acquisition, and through visual feature extraction and voiceprint feature analysis, a multi-modal fusion model is used to determine the comprehensive health index, and an environmental shielding unit is designed to isolate external interference.
It enables high-precision and rapid detection of UAV power systems in harsh environments, effectively distinguishing fault sources, reducing reliance on human experience, and improving the accuracy and consistency of detection.
Smart Images
Figure CN122126479A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a testing method and apparatus for unmanned aerial vehicle (UAV) production, belonging to the field of UAV technology. Background Technology
[0002] With the rapid development of drone technology, its applications in consumer aerial photography, agricultural plant protection, and industry inspection are becoming increasingly widespread. In the mass production of drones, the assembly quality of the power system (including motors, ESCs, and propellers) directly affects flight stability and safety. Currently, the final inspection stage on the production line mainly relies on manual test flights or simple bench tests.
[0003] However, existing detection technologies have significant limitations. On the one hand, dynamic balancing and deformation detection of propellers typically employs static measurements or offline dynamic balancing machines, which cannot simulate the aeroelastic deformation of a motor rotating at high speed. Furthermore, during full-machine power-on testing, due to the propeller's high-speed rotation, minute vibrations or cracks are difficult for the human eye to observe, and ordinary industrial cameras, limited by frame rate, struggle to capture clear transient images. On the other hand, the detection of abnormal noises in motor bearings or loose assembly often relies on workers' auditory experience or the use of simple decibel meters, which are easily affected by background noise in noisy factory environments, leading to missed detections or misjudgments. Single visual or auditory detection methods are insufficient to comprehensively characterize the health status of a power system. For example, some minute mechanical resonances may not be visually apparent but have specific characteristics in the audio frequency domain, and vice versa.
[0004] Therefore, there is an urgent need to improve existing technologies to solve the technical problems of difficulty in dynamic visual capture, poor anti-interference capability of acoustic detection, and lack of multi-dimensional data correlation judgment in the current drone production line inspection. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings and deficiencies of the existing technology and to provide a testing method and apparatus for drone production.
[0006] A testing method for unmanned aerial vehicle (UAV) production, characterized by comprising the following steps: S1. Working condition setup: Control the power system of the UAV under test to run to the preset target speed range; S2. Multidimensional data synchronous acquisition: Based on the real-time rotational speed signal of the power system, a synchronous trigger signal is generated; the synchronous trigger signal is used to control the strobe light source to illuminate the rotating component, and a freeze-frame visual image of the rotating component is acquired during the illumination; at the same time, the acoustic signal of the power system during operation is acquired. S3. Feature Extraction: Perform contour analysis on the frozen visual image to extract visual feature values reflecting physical deformation; perform spectral analysis on the acoustic signal to extract acoustic signature feature values reflecting the mechanical operating state. S4. Fusion Judgment: Input the visual feature value and voiceprint feature value into a preset multimodal fusion model to calculate the comprehensive health index of the power system, and determine whether the product is qualified based on the comprehensive health index.
[0007] Preferably, in step S2, controlling the flashing of the stroboscopic light source specifically involves: real-time monitoring of the commutation signal or back electromotive force signal of the motor to determine the real-time rotational speed of the motor; controlling the flashing frequency of the stroboscopic light source to maintain a specific harmonic relationship with the real-time rotational speed of the motor, which corresponds to the number of blades of the rotating component, thereby creating a visual persistence effect in the visual acquisition device, making the rotating component present a static image; adjusting the phase delay of the flashing signal until the predetermined observation point of the rotating component in the acquired still visual image is located at the set position of the image acquisition area.
[0008] Further, in step S3, extracting visual feature parameters includes: pre-constructing a reference contour model of the rotating component under standard qualified conditions; comparing or differentiating the contour of the rotating component in the acquired freeze-frame visual image with the reference contour model; obtaining the difference in the contour edge position between the two, and using the statistical value of the difference as the visual feature parameter to characterize the dynamic balance deviation of the propeller or the degree of blade geometric deformation.
[0009] Preferably, in step S3, extracting the acoustic fingerprint feature parameters includes: converting the collected time-domain acoustic signal into a frequency-domain acoustic fingerprint spectrum; extracting energy distribution features or cepstral coefficients in a specific frequency range from the acoustic fingerprint spectrum as the acoustic fingerprint; calculating the similarity distance between the acoustic fingerprint and the standard good product acoustic fingerprint model, and using the similarity distance as the acoustic fingerprint feature parameter to characterize the wear or assembly abnormalities of internal components of the motor.
[0010] Furthermore, step S4 specifically includes: setting visual feature thresholds and voiceprint feature thresholds; when the visual feature parameters show abnormalities but the voiceprint feature parameters are normal, generating a first type of fault prompt, pointing to an aerodynamic shape defect; when both the visual feature parameters and the voiceprint feature parameters show abnormalities, generating a second type of fault prompt, pointing to a structural assembly defect or component damage; the comprehensive health index is calculated by weighted summation based on the normalized values of the visual feature parameters and the normalized values of the voiceprint feature parameters.
[0011] The present invention also provides a testing device for UAV production, used to implement the above method, comprising: The environmental shielding unit is a closed box structure used to isolate external light and environmental noise, and has sound-absorbing and light-absorbing structures inside; The positioning and bearing unit is set inside the environmental shielding unit to fix the UAV under test and has space to allow the UAV propeller to rotate freely. The visual inspection unit includes a strobe illumination module capable of responding to high-speed pulse signals and an image acquisition module, wherein the optical axis of the strobe illumination module points to the propeller rotation area; The acoustic detection unit includes a pickup module located near the power system; The control processing unit is connected to the UAV under test, the visual detection unit, and the acoustic detection unit, respectively. The control processing unit is configured to modulate the flashing frequency of the strobe lighting module according to the real-time rotation speed of the UAV, and to perform feature extraction and fusion judgment logic.
[0012] Preferably, the positioning and bearing unit also integrates a vibration sensing module, which is connected to the control and processing unit and is used to collect mechanical vibration data of the UAV frame as an auxiliary detection parameter for fusion judgment.
[0013] Preferably, the environmental shielding unit is further provided with a visual calibration reference object, which is located within the field of view of the image acquisition module and is used to perform distortion correction or size reference calibration on the image acquisition module during the detection process.
[0014] The beneficial effects of this invention are as follows: First, this invention utilizes hard-synchronized stroboscopic vision technology, which can capture clear, still images of high-speed rotating (e.g., above 10,000 rpm) propellers on ordinary industrial cameras at extremely low cost, directly measuring blade tip trajectory deviation and blade torsional deformation under dynamic operating conditions. This overcomes the limitation of traditional static detection in failing to detect deformation problems caused by centrifugal force, and compared to expensive high-speed camera solutions, the data processing volume is significantly reduced, the detection cycle is faster, and it is suitable for large-scale production line applications.
[0015] Secondly, this invention constructs a judgment logic for audiovisual fusion. By associating minute differences in visual contours with audio frequency characteristics, fault sources can be effectively distinguished. For example, simple blade deformation will cause visual deviation, but the sound change is mainly in low-frequency wind noise, while motor shaft bending will cause visual deviation and be accompanied by high-frequency periodic mechanical noise. This logic can automatically output fault classification (aerodynamic defect vs. structural defect), providing clear guidance for production line rework and reducing reliance on human experience.
[0016] Finally, the device designed in this invention employs an environmental shielding unit that integrates sound absorption and light-absorbing structures, effectively isolating the complex background light and noise of the production workshop. Combined with internal visual calibration references, it ensures the consistency of the testing benchmark for each instance, thereby guaranteeing a high signal-to-noise ratio and high repeatability of the testing data in harsh industrial environments. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, obtaining other drawings based on these drawings without creative effort still falls within the scope of the present invention.
[0018] Figure 1 A schematic flowchart of a testing method for unmanned aerial vehicle (UAV) production provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the timing logic for flicker control and image acquisition in an embodiment of the present invention; Figure 3 This is a block diagram illustrating the principle of the multimodal fusion determination logic in an embodiment of the present invention. Figure 4 This is a structural diagram of a testing device for UAV production provided in an embodiment of the present invention; Figure 5 An internal structural diagram of a testing device for UAV production provided in an embodiment of the present invention; In the diagram: 10. Environmental shielding unit; 11. Sound-absorbing material; 12. Matte coating; 20. Positioning and bearing unit; 21. Vibration sensing module; 30. Visual inspection unit; 31. Strobe lighting module; 32. Image acquisition module; 40. Acoustic inspection unit; 50. Control and processing unit; 60. Visual calibration reference. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.
[0020] It should be noted that all uses of "first" and "second" in the embodiments of the present invention are for the purpose of distinguishing two entities or parameters with the same name but different names. It is clear that "first" and "second" are only for the convenience of expression and should not be construed as limiting the embodiments of the present invention. Subsequent embodiments will not explain this in detail.
[0021] The directional and positional terms used in this invention, such as "up," "down," "front," "back," "left," "right," "inner," "outer," "top," "bottom," and "side," are merely for reference to the accompanying drawings. Therefore, the directional and positional terms used are for illustrating and understanding this invention, and not for limiting the scope of protection of this invention.
[0022] To address the shortcomings of existing drone production line inspection methods, such as limited options and difficulty in capturing dynamic defects, this application provides a drone production inspection method to improve upon these issues. Figure 1-3 As shown, it includes steps S100-S400.
[0023] S100, Operating Condition Setup: Control the power system of the UAV under test to operate within the preset target speed range.
[0024] The control processing unit sends PWM signals to the drone motors via an electronic speed controller (ESC) to accelerate the motors to typical cruise speeds or maximum load speeds. Specifically, to expose potential mechanical problems, a stepped speed curve is usually set, such as climbing from idle speed to 50% throttle and then instantly accelerating to 100% throttle, maintaining stability throughout the process for subsequent data collection.
[0025] S200, Multi-dimensional Data Synchronous Acquisition: Based on the real-time rotational speed signal of the power system, a synchronous trigger signal is generated; the synchronous trigger signal is used to control the strobe light source to illuminate the rotating component, and a freeze-frame visual image of the rotating component is acquired during the illumination; at the same time, the acoustic signal of the power system during operation is acquired.
[0026] The core of this step lies in phase locking. Specifically, the system does not rely on external photogates, but directly reads the commutation signal (for brushless motors) or the zero-crossing point of the back EMF in the motor drive circuit. Once the speed is detected to be stable, the control processing unit generates a pulse sequence that is strictly synchronized with the speed to trigger a high-brightness LED strobe light.
[0027] S300, Feature Extraction: Perform contour analysis on the frozen visual image to extract visual feature values reflecting physical deformation; perform spectral analysis on the acoustic signal to extract acoustic signature feature values reflecting the mechanical operating state.
[0028] Among them, visual feature values mainly refer to the positional fluctuation of the blade tip in the image coordinate system; acoustic feature values refer to the energy density of a specific frequency band (such as the characteristic frequency of bearing failure).
[0029] S400, Fusion Judgment: Input the visual feature value and voiceprint feature value into a preset multimodal fusion model to calculate the comprehensive health index of the power system, and determine whether the product is qualified based on the comprehensive health index.
[0030] The present invention further proposes that, in step S200, controlling the flashing of the stroboscopic light source specifically involves: real-time monitoring of the commutation signal or back electromotive force signal of the motor to determine the real-time rotational speed of the motor; controlling the flashing frequency of the stroboscopic light source to maintain a specific harmonic relationship with the real-time rotational speed of the motor, wherein the harmonic relationship corresponds to the number of blades of the rotating component, thereby forming a visual persistence effect in the visual acquisition device, making the rotating component present a static image; adjusting the phase delay of the flashing signal until the predetermined observation point of the rotating component in the acquired still visual image is located at the set position of the image acquisition area.
[0031] The frequency multiplication relationship refers to the flicker frequency of a stroboscopic light source if the motor speed is n (rpm) and the propeller has N blades. (Hz) can be set to the following relationship: Where m is an integer multiple of the number of blades N (e.g., m=N or m=1). Specifically, through this hard synchronization mechanism, regardless of fluctuations in motor speed, the strobe light always strikes the propeller at the exact moment it reaches the same angle. Furthermore, the phase delay time... The adjustment logic can be expressed as: Where θ is the target phase angle of the target observation point during the rotation period. Through fine-tuning... A stationary propeller can be rotated within the image frame until the blade tip is precisely within the camera's optimal focus and measurement area, thus achieving high-precision point-to-point observation. As an alternative embodiment, those skilled in the art can use a photoelectric sensor to detect the propeller's passage time to generate a trigger signal, but this is significantly affected by dust and is less stable than electrical signal feedback.
[0032] The present invention further proposes that, in step S300, the extraction of visual feature parameters includes: pre-constructing a reference contour model of the rotating component under standard qualified conditions; performing overlap comparison or differential processing on the contour of the rotating component in the acquired freeze-frame visual image and the reference contour model; obtaining the difference in the contour edge position between the two, and using the statistical value of the difference as the visual feature parameter to characterize the dynamic balance deviation of the propeller or the degree of blade geometric deformation.
[0033] The reference contour model is a mask generated based on CAD design drawings or standard gold sample acquisition. Specifically, in the image processing algorithm, the system binarizes the real-time acquired "static" blade image and extracts its Canny edges. Assume the set of points on the standard contour is... The set of points on the contour is collected in real time. The difference in the position of the contour edges, ΔP, can be quantified by calculating the maximum Hausdorff distance between them: If ΔP exceeds a preset threshold, or if the difference is mainly concentrated in the vertical direction of the blade tip, it indicates that the blade is warped or experiencing up-and-down flapping due to dynamic imbalance. This method can detect dynamic deformation at the 0.5mm level, which is impossible with static caliper measurements.
[0034] The present invention further proposes that, in step S300, extracting acoustic fingerprint feature parameters includes: converting the acquired time-domain acoustic signal into a frequency-domain acoustic fingerprint spectrum; extracting energy distribution features or cepstral coefficients in a specific frequency range from the acoustic fingerprint spectrum as acoustic fingerprints; calculating the similarity distance between the acoustic fingerprint and a standard good acoustic fingerprint model, and using the similarity distance as the acoustic fingerprint feature parameter to characterize the wear or assembly abnormalities of internal components of the motor.
[0035] The "voiceprint fingerprint" utilizes Mel-frequency cepstral coefficients (MFCC). Specifically, the sound of a motor not only contains the fundamental frequency but also rich higher harmonics. When the bearing balls are worn or lack lubrication, abnormal broadband noise or modulation sidebands will appear in the high-frequency range (e.g., 5kHz-10kHz). The system converts the sound signal into a sound spectrum using a Fast Fourier Transform (FFT). Assuming the extracted MFCC feature vector is... The feature vector of a standard good product is Then the voiceprint similarity distance Euclidean distance can be used for calculation: Where k is the dimension of the feature vector. This non-contact auscultation can penetrate the casing to detect internal problems.
[0036] The present invention further proposes that step S400 specifically includes: setting visual feature thresholds and voiceprint feature thresholds; when the visual feature parameters show abnormality but the voiceprint feature parameters are normal, generating a first type of fault prompt, pointing to an aerodynamic shape defect; when both the visual feature parameters and the voiceprint feature parameters show abnormality, generating a second type of fault prompt, pointing to a structural assembly defect or component damage; the comprehensive health index is calculated by weighted summation based on the normalized values of the visual feature parameters and the normalized values of the voiceprint feature parameters.
[0037] This classification logic significantly improves the efficiency of fault tracing. Specifically, as a particular implementation method, the comprehensive health index H can be calculated using the following weighted formula: in, and These are the maximum permissible visual deformation threshold and the maximum voiceprint difference threshold, respectively, with α and β being preset weighting coefficients (e.g., α=0.6, β=0.4). If H is less than the preset pass / fail score (e.g., 0.8), the product is deemed unqualified. This makes the quality inspection result not just "qualified / unqualified," but a quantitative quality score, which helps in process improvement.
[0038] The present invention further proposes, such as Figure 4-5 As shown, a testing device for drone production, used to implement the above method, includes: an environmental shielding unit 10, which is a closed box structure used to isolate external light and environmental noise, and has sound-absorbing and light-absorbing structures inside; a positioning and bearing unit 20, which is disposed within the environmental shielding unit 10, used to fix the drone under test, and has space for the drone propeller to rotate freely; a visual inspection unit 30, including a strobe illumination module 31 capable of responding to high-speed pulse signals and an image acquisition module 32, wherein the optical axis of the strobe illumination module 31 points to the propeller rotation area; an acoustic inspection unit 40, including a sound pickup module disposed near the power system; and a control processing unit 50, which is connected to the drone under test, the visual inspection unit 30, and the acoustic inspection unit 40 respectively; the control processing unit 50 is configured to modulate the flashing frequency of the strobe illumination module 31 according to the real-time rotation speed of the drone, and to perform feature extraction and fusion judgment logic.
[0039] The ambient light shielding unit 10 features an internal matte structure (made of black flocked cloth or matte paint) that prevents multiple reflections of strobe light within the enclosure, thus preventing image glare and ensuring the accuracy of contour extraction by the visual algorithm. The sound-absorbing structure (made of corrugated sponge or other materials) attenuates background noise from the external production line. This device creates a standardized black-box testing environment, ensuring that each drone is tested under completely consistent lighting and acoustic boundary conditions, eliminating misjudgments caused by environmental factors.
[0040] The present invention further proposes that the positioning and bearing unit 20 is also integrated with a vibration sensing module 21. The vibration sensing module 21 is connected to the control and processing unit 50 and is used to collect mechanical vibration data of the UAV frame as auxiliary detection parameters to participate in the fusion judgment.
[0041] The positioning and support unit uses a cylinder-driven gripper to fix the drone. Under the control of the control processing unit, the cylinder can open or release the gripper. It is also highly compatible with drones of different bodies and sizes.
[0042] The vibration sensing module 21 is typically a triaxial accelerometer (MEMS). Specifically, contact-type vibration sensors are most sensitive to low-frequency resonance of the entire robotic arm. Introducing data from this dimension can create a three-dimensional detection network integrating sight, sound, and touch. For example, when resonance occurs, visual deviation may be small, and sound may be drowned out, but the vibration sensor will measure extremely high G-values, thus filling in the detection blind spots.
[0043] The vibration sensing module is located at the center of the gripper, and the gripper's stroke is controlled to be unaffected by the gripper, while still allowing the mobile phone data to be transmitted in contact with the drone body.
[0044] The present invention further proposes that the environmental shielding unit 10 is also provided with a visual calibration reference object 60, which is located within the field of view of the image acquisition module 32 and is used to perform distortion correction or size reference calibration on the image acquisition module 32 during the detection process.
[0045] The visual calibration reference 60 can be a checkerboard or circular array calibration plate. Specifically, since industrial camera lenses have distortion and the installation position may shift slightly due to vibration, by setting a fixed calibration object in one corner of the field of view, the algorithm can automatically calculate the conversion ratio between pixels and physical size (millimeters) before each detection and correct image distortion, ensuring that the measured deformation is the true physical value, rather than an error caused by lens distortion.
[0046] Working principle: Once the UAV under test is fixed on the positioning support unit 20 and started, the control processing unit 50 reads the real-time rotational speed of the motor. Assuming the motor speed is 6000 RPM, the controller drives the strobe lighting module 31 to flash at a frequency of 100Hz (corresponding to a two-bladed propeller), with an extremely narrow pulse width (e.g., 50 microseconds). At this time, the image acquisition module 32 captures a clear, still image of the propeller at the same angular position in each frame. Algorithm analysis of this image reveals that the propeller tip contour is offset outward by 2 pixels (corresponding to 0.5mm) compared to the standard model. Simultaneously, the acoustic detection unit 40 collects sound signals and analyzes them, finding an abnormal peak at 2kHz. The multimodal fusion model integrates these two features, calculates a health index below a threshold, and determines it as "slight bending of the motor shaft." The device then issues an alarm and stops the test.
[0047] By freezing the shape of rotating parts through hard synchronization of motor speed and strobe, and then correlating this dynamic shape characteristic with acoustic operation characteristics, it can be used for online quality inspection of UAV production lines.
[0048] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.
[0049] While the invention has been described with reference to several specific embodiments, it should be understood that the invention is not limited to the disclosed specific embodiments. The invention is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.
Claims
1. A testing method for unmanned aerial vehicle (UAV) production, characterized in that: Includes the following steps: S1. Working condition setup: Control the power system of the UAV under test to run to the preset target speed range; S2. Multidimensional data synchronous acquisition: Based on the real-time rotational speed signal of the power system, a synchronous trigger signal is generated; the synchronous trigger signal is used to control the strobe light source to illuminate the rotating component, and a freeze-frame visual image of the rotating component is acquired during the illumination; at the same time, the acoustic signal of the power system during operation is acquired. S3. Feature extraction: Perform contour analysis on the frozen visual image to extract visual feature values that reflect physical deformation; The acoustic signal is subjected to spectral analysis to extract acoustic signature features that reflect the mechanical operating state; S4. Fusion Judgment: Input the visual feature value and voiceprint feature value into a preset multimodal fusion model to calculate the comprehensive health index of the power system, and determine whether the product is qualified based on the comprehensive health index.
2. The detection method for human-machine production as described in claim 1, characterized in that: In step S2, the specific steps for controlling the flickering of the stroboscopic light source are as follows: Real-time monitoring of the motor's commutation signal or back electromotive force signal determines the motor's real-time speed; The flashing frequency of the stroboscopic light source is controlled to maintain a specific frequency doubling relationship with the real-time rotational speed of the motor. This frequency doubling relationship corresponds to the number of blades of the rotating component, thereby creating a visual persistence effect in the visual acquisition device, making the rotating component present a static image. Adjust the phase delay of the flash signal until the predetermined observation point of the rotating component in the captured freeze-frame visual image is located at the set position in the image acquisition area.
3. The detection method for human-machine production as described in claim 1, characterized in that: In step S3, the extraction of visual feature parameters includes: A reference profile model of the rotating component under standard qualified conditions is pre-constructed; The contours of the rotating parts in the captured freeze-frame visual images are compared or differentially processed with the reference contour model. The difference in the position of the contour edges between the two is obtained, and the statistical value of the difference is used as the visual feature parameter to characterize the dynamic balance deviation of the propeller or the degree of blade geometric deformation.
4. The detection method for human-machine production as described in claim 1, characterized in that: In step S3, the extraction of voiceprint feature parameters includes: The acquired time-domain acoustic signal is converted into a frequency-domain acoustic spectrum. In the acoustic signature spectrum, energy distribution features or cepstral coefficients in a specific frequency range are extracted as acoustic fingerprints; Calculate the similarity distance between the voiceprint fingerprint and the standard good product voiceprint model, and use the similarity distance as the voiceprint feature parameter to characterize the wear or assembly abnormality of the internal components of the motor.
5. The detection method for human-machine production as described in claim 1, characterized in that: The S4 step specifically includes: Set visual feature thresholds and voiceprint feature thresholds; When the visual feature parameters show abnormalities but the acoustic feature parameters are normal, a first-type fault prompt is generated, pointing to an aerodynamic shape defect. When both visual feature parameters and voiceprint feature parameters show abnormalities, a second type of fault indication is generated, pointing to structural assembly defects or component damage. The comprehensive health index is calculated by weighted summation of the normalized values of visual feature parameters and voiceprint feature parameters.
6. A testing device for unmanned aerial vehicle (UAV) production, used to implement the method described in any one of claims 1 to 5, characterized in that, include: The environmental shielding unit is a closed box structure used to isolate external light and environmental noise, and has sound-absorbing and light-absorbing structures inside; The positioning and bearing unit is set inside the environmental shielding unit to fix the UAV under test and has space to allow the UAV propeller to rotate freely. The visual inspection unit includes a strobe illumination module capable of responding to high-speed pulse signals and an image acquisition module, wherein the optical axis of the strobe illumination module points to the propeller rotation area; The acoustic detection unit includes a pickup module located near the power system; The control processing unit is connected to the UAV under test, the visual detection unit, and the acoustic detection unit, respectively. The control processing unit is configured to modulate the flashing frequency of the strobe lighting module according to the real-time rotation speed of the UAV, and to perform feature extraction and fusion judgment logic.
7. The testing device for UAV production as described in claim 6, characterized in that: The positioning and bearing unit also integrates a vibration sensing module, which is connected to the control and processing unit and is used to collect mechanical vibration data of the UAV frame as an auxiliary detection parameter for fusion judgment.
8. The testing device for UAV production as described in claim 6, characterized in that: The environmental shielding unit is also equipped with a visual calibration reference object, which is located within the field of view of the image acquisition module and is used to perform distortion correction or size reference calibration of the image acquisition module during the detection process.