Portable AI sound spectrum analyzer device and use method thereof

Abstract

The invention relates to the technical field of acoustic detection and signal processing equipment, in particular to a portable AI sound spectrum analyzer device and a use method thereof. The device comprises a box body, an embedded control panel, a tool assembly, a power supply module and an acquisition connection module. The tool assembly comprises a quick-change profiling support and a miniature vibration sensor, and a sound collector support supports three-dimensional adjustment. The embedded control panel is integrated with an AI sound spectrum analysis system, and end side noise reduction can be carried out on high-fidelity audio and vibration signals. The beneficial effects of the invention are that: through the integration of the modular tool, the multi-degree-of-freedom sound collection structure, the high-fidelity collection link and the embedded AI reasoning engine, the problems that equipment cannot be moved, the tool cannot be reused, the noise suppression capability is weak and the analysis intelligence degree is low in the acoustic detection of multi-class products in the industrial field are solved.

Description

Technical Field

[0001] This invention relates to the field of acoustic detection and signal processing equipment technology, and in particular to a portable AI acoustic spectrum analyzer device and its usage method. Background Technology

[0002] With the deepening integration of artificial intelligence and acoustic detection technology, portable acoustic spectrum analysis equipment is showing broad application prospects in fields such as industrial product quality inspection, equipment fault diagnosis, and environmental noise monitoring. While various sound detection or analysis devices exist in the current technology, most focus on specific scenarios (such as communication interference identification, voice keyword detection, or abnormal audio classification), lacking a portable solution that integrates high-fidelity acquisition and embedded AI intelligent analysis for rapid, accurate, and adaptable detection of the acoustic performance of industrial products. Especially when dealing with products of different structures, sizes, and sound characteristics, existing equipment often struggles to achieve efficient integrated operation of sound acquisition, localization, and intelligent analysis.

[0003] Existing technology, patent CN104581608B, discloses a sound detection circuit system published on April 6, 2018. This patent provides a circuit system for testing the sound output performance of electronic products, evaluating the product's sound performance under different loads through an audio interface, load switching, and a central control unit. However, this system is essentially a fixed benchtop testing device, relying on wired connections and preset interfaces, lacking portability; furthermore, its core function is limited to measuring electroacoustic performance parameters, without integrating spectral visualization, AI model inference, or adaptive tooling support for non-standard products, thus failing to meet the needs of rapid on-site deployment and acoustic characteristic analysis of multiple product categories.

[0004] Therefore, a portable AI acoustic spectrum analyzer device needs to be designed to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a portable AI acoustic spectrum analyzer device to overcome the aforementioned shortcomings of the existing technology.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A portable AI acoustic spectrum analyzer device includes a housing body, an embedded control board placed inside the housing body, a tooling assembly disposed on the side of the embedded control board, a power supply module for supplying power to the product under test and driving the operation of the whole machine, and an acquisition connection module for acquiring sound and vibration signals. A sound collector bracket is disposed on one side of the tooling assembly. The embedded control board integrates an artificial intelligence-based acoustic spectrum analysis and processing system and a network communication module. The embedded control board is electrically connected to the power module and the acquisition and connection module via wiring harnesses. The acquisition and connection module includes a network port, a data transmission interface, a dedicated audio interface for the sound collector, and a dedicated BNC interface for the vibration sensor, all located on the front side panel of the enclosure.

[0007] Preferably, a voltage adjustment knob is provided on one side of the embedded control board, and the voltage adjustment knob is connected to the adjustable voltage regulator circuit in the power supply module through a wire.

[0008] Preferably, the power module includes a power supply body, a power support fixedly installed on the inner wall of the housing body, and four sets of standard power interfaces leading out from the power supply body and passing through the side wall of the housing body. Each set of power interfaces is connected to the power supply terminal of the product under test through a shielded wire harness.

[0009] Preferably, the network port is connected to the network controller of the embedded control board via a gigabit Ethernet PHY chip; the data transmission interface is connected to the data bus of the embedded control board via the controller; the dedicated audio interface for the sound collector is connected to the digital signal processor of the embedded control board via a high-fidelity audio codec; and the dedicated BNC interface for the vibration sensor is connected to the analog input channel of the embedded control board via an isolation amplifier circuit.

[0010] Preferably, the tooling assembly includes a tooling base fixedly mounted on the bottom plate of the housing body, a contouring bracket detachably snapped onto the upper surface of the tooling base, and a miniature vibration sensor fixed to the top of the contouring bracket by threaded fasteners; the contour of the contouring bracket is customized according to the outer contour of the product to be tested, and its bottom is provided with a positioning pin hole, which cooperates with the positioning pin provided on the tooling base.

[0011] Preferably, the microphone bracket includes an axial bracket vertically fixed to the bottom plate of the housing body, a transverse bracket hinged to the top of the axial bracket, and a microphone mounting base connected to the end of the transverse bracket via a universal ball joint; the microphone is screwed into the microphone mounting base by threads and connected to the microphone's dedicated audio interface via an audio cable; the universal ball joint includes a locking knob.

[0012] Preferably, the tooling base is provided with a quick-clamping mechanism for locking the contour bracket onto the tooling base.

[0013] The method of using a portable AI acoustic spectrum analyzer device includes the following steps: S1. Tooling assembly: Take the contour bracket out of the box body, align the positioning pin hole at its bottom with the positioning pin on the tooling base, and lock it in place using the quick-clamp mechanism; fix the micro vibration sensor in the mounting hole reserved at the top of the contour bracket using M3 screws; S2. Sound collector positioning: unfold the horizontal bracket and rotate it around the hinge axis to a horizontal position, loosen the locking knob of the universal ball joint, move the sound collector to a distance of 5-20cm from the sound-emitting part of the product to be tested, and adjust its direction to face the sound-emitting surface to be tested, and then tighten the locking knob. S3. Cable connection: Insert one end of the power supply cable of the product under test into the power interface of the corresponding specification, and connect the other end to the power input terminal of the product under test; insert the audio cable of the sound collector into the dedicated audio interface of the sound collector; insert the signal cable of the miniature vibration sensor into the dedicated BNC interface of the vibration sensor. S4. Start Acquisition: The main control program of the embedded control board is started, and the high-fidelity audio acquisition channel and vibration signal acquisition channel are started; the sound collector converts physical sound waves into analog electrical signals, which are then converted into 24-bit / 96kHz digital audio streams by the high-fidelity audio codec; the analog voltage signal output by the miniature vibration sensor is conditioned by the isolation amplifier circuit and then digitized by the ADC module of the embedded control board at a sampling rate of 1MHz; S5. Signal Processing: The embedded control board performs the following operations: (a) bandpass filtering on the digital audio stream, with the cutoff frequency set to 100Hz–10kHz; (b) background noise separation of the audio signal using a deep noise reduction network based on the U-Net architecture; (c) framing the audio signal with a frame length of 25ms and a frame shift of 10ms, and applying a Hamming window to each frame; (d) calculating the short-time Fourier transform of each frame, generating an amplitude spectrum, and inputting it into a pre-trained convolutional neural network for abnormal voiceprint recognition; (e) simultaneously performing wavelet packet decomposition on the vibration signal, extracting energy entropy features, fusing them with acoustic features, and inputting them into a multimodal classification model. S6. Result Output: The embedded control board outputs the analysis results in the following ways: (i) drawing real-time spectrograms, vibration time-domain waveforms and abnormal event markers on the local display screen; (ii) exporting a structured detection report through the data transmission interface; (iii) if an abnormal sound pattern is detected, triggering a buzzer alarm and uploading alarm logs to the remote server through the network port.

[0014] Preferably, in step S5, the training dataset of the deep noise reduction network includes more than 1,000 normal sound samples of products collected under different industrial noise environments, and is optimized using an adversarial training strategy.

[0015] The beneficial effects of this invention are: This technical solution solves the problems of poor equipment portability, low tooling adaptation efficiency, weak noise suppression capability and low level of intelligent analysis in the acoustic testing of multiple product categories in industrial sites by integrating modular tooling, multi-degree-of-freedom sound acquisition structure, high-fidelity acquisition link and embedded AI inference engine. The tooling components use a quick-change contour bracket and a standardized base. For different products, only the contour bracket needs to be replaced, without the need to recalibrate the whole machine, which significantly shortens the on-site deployment time. The microphone bracket has a three-level adjustment structure of axial, lateral and universal, which enables the microphone to be precisely positioned in three-dimensional space, ensuring that it can still be aligned with the best pickup point even under complex product shapes, thereby improving the signal-to-noise ratio. The embedded control board's built-in deep noise reduction network and multimodal analysis model enable edge-side intelligent inference without the need for external server support, avoiding data upload delays and privacy leak risks. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of a portable AI acoustic spectrum analyzer device according to the present invention; Figure 2 This is a schematic diagram of the test state of a portable AI acoustic spectrum analyzer device according to the present invention; Figure 3 This is a second-view schematic diagram of a portable AI acoustic spectrum analyzer device according to the present invention; In the diagram: 1. Cabinet body; 2. Embedded control board; 3. Tooling base; 4. Contouring bracket; 5. Miniature vibration sensor; 6. Sound collector; 7. Axial bracket; 8. Lateral bracket; 11. Voltage adjustment knob; 12. Power interface; 13. Dedicated audio interface for sound collector; 14. Dedicated BNC interface for vibration sensor; 15. Network port; 16. Data transmission interface; 17. Power module. Detailed Implementation

[0017] Reference Figures 1 to 3 A portable AI acoustic spectrum analyzer device includes a housing body 1, an embedded control board 2 placed inside the housing body 1, a tooling assembly disposed on the side of the embedded control board 2, a power supply module 17 for supplying power to the product under test and driving the operation of the whole machine, and an acquisition connection module for acquiring sound and vibration signals.

[0018] The embedded control board integrates an AI-based acoustic spectrum analysis and processing system and a network communication module. The embedded control board 2 is electrically connected to the power module 17 and the acquisition and connection module via wiring harnesses. As the core processing unit of the device, the embedded control board 2 integrates a multi-channel synchronous data acquisition module, a real-time digital signal processing module, and an embedded AI inference engine.

[0019] A voltage adjustment knob 11 is provided on one side of the embedded control board 2. The voltage adjustment knob 11 is connected to the adjustable voltage regulator circuit in the power module 17 through a wire. The power module 17 includes a power supply body, a power support fixedly installed on the inner wall of the housing body, and four sets of standard power interfaces 12 leading out from the power supply body and passing through the side wall of the housing body 1. Each set of power interfaces 12 is connected to the power supply terminal corresponding to the product under test through a shielded wire harness. By integrating and isolating the power module 17, electromagnetic interference from internal cables and its own operating noise are reduced, ultimately achieving low operating noise of the entire clamping mechanism, avoiding signal contamination, and providing a consistent and pure acoustic signal source for AI quality inspection.

[0020] The acquisition and connection module includes a network port 15, a data transmission interface 16, a dedicated audio interface 13 for the sound collector, and a dedicated BNC interface 14 for the vibration sensor, all located on the front panel of the housing 1. The network port 15 is connected to the network controller of the embedded control board 2 via a gigabit Ethernet PHY chip. The data transmission interface 16 is connected to the data bus of the embedded control board 2 via the controller. The dedicated audio interface 13 for the sound collector is connected to the digital signal processor of the embedded control board 2 via a high-fidelity audio codec. The dedicated BNC interface 14 for the vibration sensor is connected to the analog input channel of the embedded control board 2 via an isolation amplifier circuit. The tooling assembly includes a tooling base 3 fixedly installed on the bottom plate of the housing body 1, a contouring bracket 4 detachably snapped onto the upper surface of the tooling base 3, and a miniature vibration sensor 5 fixed to the top of the contouring bracket 4 by threaded fasteners; the contour of the contouring bracket 4 is customized according to the shape contour of the product to be tested, and its bottom is provided with a positioning pin hole, which cooperates with the positioning pin provided on the tooling base 3. A microphone bracket is provided on one side of the tooling assembly. The microphone bracket includes an axial bracket 7 that is vertically fixed to the bottom plate of the housing body 1, a transverse bracket 8 that is hinged to the top of the axial bracket 7, and a microphone mounting base that is connected to the end of the transverse bracket 8 via a universal ball joint. The microphone 6 is screwed into the microphone mounting base and connected to the microphone-specific audio interface 13 via an audio cable. The universal ball joint includes a locking knob.

[0021] The main body of the enclosure 1 is made of engineering plastic with an IP54 protection rating. It has a rectangular structure, a handle on the top, and casters and a retractable rod on the bottom for easy carrying by a single person and on-site deployment. The interior of the main body of the enclosure 1 forms a closed chamber to accommodate the various functional modules, and multiple interfaces and adjustment components are set on its front panel.

[0022] The embedded control board 2 is fixedly installed inside the main body 1 on the left side and secured to the inner wall bracket with screws. The embedded control board 2 uses a quad-core ARM Cortex-A72 processor as the main control chip, is equipped with 8GB LPDDR4 memory and 256GB eMMC storage, and runs a Linux-based real-time operating system. The embedded control board 2 integrates a digital signal processor, network controller, controller, high-fidelity audio codec, analog input channel, and ADC module. The embedded control board 2 is electrically connected to the power module 17 and the data acquisition module via multiple shielded wire harnesses, forming the core of the entire device's control and data processing. The power module 17 includes a power supply body and a power supply bracket. The power supply bracket is a metal stamping part, fixed to the inner right side wall of the housing body 1 by bolts. The power supply body is snapped into the power supply bracket and locked by a limiting baffle. The power supply body contains an adjustable voltage regulator circuit, whose output terminals lead out several sets of standard power interfaces 12. Each set of power interfaces 12 is a red and black banana-shaped jack, passing through the right side wall of the housing body 1 and exposed on the panel surface. The four sets of power interfaces 12 are connected to different output channels of the adjustable voltage regulator circuit through independent shielded wire harnesses, achieving multi-channel isolated DC power supply. The voltage adjustment knob 11 is installed next to the embedded control board. Its rotation axis is connected to the control terminal of the adjustable voltage regulator circuit through a wire, used to manually set the output voltage value. The adjustment range is 0–24V DC, with a step accuracy of 0.1V. The data acquisition module is located in the middle of the front panel of the enclosure 1, and includes a network port 15, a data transmission interface 16, a dedicated audio interface 13 for the audio collector, and a dedicated BNC interface 14 for the vibration sensor. The network port 15 is an RJ45 Gigabit Ethernet interface, which is internally connected to the network controller on the embedded control board 2 via a Gigabit Ethernet PHY chip. The data transmission interface 16 is connected to the data bus of the embedded control board 2 via the controller. The dedicated audio interface 13 for the audio collector is an XLR three-pin connector, which is internally connected to the digital signal processor of the embedded control board 2 via a high-fidelity audio codec. The dedicated BNC interface 14 for the vibration sensor is connected to the input of the isolation amplifier circuit via a coaxial cable. The output of the isolation amplifier circuit is connected to the analog input channel of the embedded control board 2, which is equipped with a 16-bit high-precision ADC module supporting a sampling rate of up to 1MHz.

[0023] The tooling assembly is installed inside the housing body 1, including a tooling base 3, a contour bracket 4, a miniature vibration sensor 5, and a sound collector bracket. The tooling base 3 is an aluminum alloy casting with grooves inside the housing body 1 for placing the tooling components. Specifically, the upper surface of the tooling base has two locating pins arranged diagonally. The pins are cylindrical stainless steel parts, 8mm high and 6mm in diameter. The contour bracket 4 is CNC machined from 6061-T6 aluminum alloy and anodized. Its lower surface has locating pin holes matching the locating pins, with a diameter of 6.1mm and a depth of 10mm. When the contour bracket 4 is placed on the tooling base 3, the locating pins are inserted into the locating pin holes, achieving precise horizontal positioning. The upper surface of the contour bracket 4 has custom grooves or support surfaces based on the outline of the product to be tested, and an M3 threaded hole is pre-drilled at the top for mounting the miniature vibration sensor 5.

[0024] In other implementation cases, a quick-clamp mechanism is also provided. This structure is an existing fixture, which is specifically installed on one side edge of the tooling base 3. The end of its pressure arm is equipped with a rubber pressure head. When the contour bracket 4 is in place, the handle of the quick-clamp mechanism 18 is turned, the pressure arm presses down and locks the side flange of the contour bracket 4 to prevent it from shifting during the test.

[0025] The microphone bracket includes an axial bracket 7, a transverse bracket 8, a universal ball joint, and a locking knob. The axial bracket 7 is a 12mm diameter stainless steel column, its bottom end welded to the base plate of the housing body 1, and its top end has a hinge hole. The transverse bracket 8 is a 200mm long aluminum alloy rod, one end of which is connected to the hinge hole at the top of the axial bracket 7 via a pin, forming a hinge structure that can rotate around a horizontal axis, with a rotation angle range of 0°–90°. The other end of the transverse bracket 8 is fixedly connected to the housing of the universal ball joint. Inside the universal ball joint is a 15mm diameter steel ball head, one end of which is fixedly connected to the microphone mounting base, and the other end is embedded in the joint housing, allowing free rotation in three-dimensional space. The locking knob is a threaded knob that screws into the side wall of the universal ball joint housing. When tightened, it presses against the ball head, creating friction between it and the inner wall of the housing, thereby fixing the spatial orientation of the microphone 6. The microphone 6 is a 1 / 2-inch prepolarized condenser microphone with a standard UNC-2A thread at its tail. It is screwed into the corresponding screw hole of the microphone mounting base. The output of the microphone 6 is connected to the dedicated audio interface 13 of the microphone via a 3-core shielded audio cable.

[0026] In actual use, the operator first selects the corresponding contour bracket 4 according to the type of product to be tested. Taking a certain model of small fan as an example, its shape is cylindrical, and the upper surface of the contour bracket 4 has a matching positioning groove. The operator removes the contour bracket 4 from the housing body 1, aligns the two positioning pin holes at its bottom with the positioning pins on the tooling base 3, and lowers it vertically so that the positioning pins are fully inserted into the pin holes. Then, the miniature vibration sensor 5 is screwed into the threaded hole at the top of the contour bracket 4 using an M3 screw, ensuring that the bottom surface of the sensor is in close contact with the outer shell of the product to be tested; Next, unfold the microphone bracket: rotate the horizontal bracket 8 upwards around the hinge axis at the top of the axial bracket 7 to a horizontal position, at which point the horizontal bracket 8 and the axial bracket 7 form a 90° angle. Loosen the locking knob on the universal ball joint, and hold the microphone 6 to adjust its spatial position along the ball joint, so that the pickup diaphragm of the microphone 6 is facing the sound-emitting part of the fan being tested, maintaining a distance of 5–20 cm. After adjustment, tighten the locking knob to fix the position of the microphone 6.

[0027] During the cable connection phase, insert one end of the power cord of the motor under test into the corresponding power interface 12 (e.g., red jack is positive, black is negative), and connect the other end to the motor's power input terminal. Insert the audio cable of the audio collector 6 into the dedicated audio interface 13 of the audio collector, and insert the signal cable of the miniature vibration sensor 5 into the dedicated BNC interface 14 of the vibration sensor. After all connections are completed, start the main control program of the embedded control board 2, and the system will automatically initialize each acquisition channel.

[0028] After the acquisition is initiated, the audio collector 6 converts the received physical sound waves into analog electrical signals. These signals are then input to a high-fidelity audio codec via the dedicated audio interface 13, where they undergo 24-bit / 96kHz analog-to-digital conversion to generate a digital audio stream, which is then sent to the digital signal processor of the embedded control board 2. Simultaneously, the analog voltage signal output by the miniature vibration sensor 5 enters an isolation amplifier circuit via the BNC interface 14. This circuit performs common-mode rejection and amplitude conditioning on the signal, outputting a voltage signal adapted to the ADC input range. This voltage signal is then digitized by the ADC module of the embedded control board 2 at a sampling rate of 1MHz.

[0029] In the signal processing stage, the embedded control board 2 first applies a 100Hz–10kHz digital bandpass filter to the digital audio stream to filter out mechanical vibration interference below 10Hz and electromagnetic noise above 10kHz. Then, it calls a pre-loaded deep noise reduction network model based on the U-Net architecture. This model, during its training phase, uses normal sound samples from over 1000 types of industrial products under different background noise levels (such as fans, compressors, and assembly lines), and employs an adversarial training strategy for optimization, effectively separating the target sound source from environmental noise. The denoised audio signal is then segmented into short-time frames with a 25ms frame length and a 10ms frame shift. Each frame is multiplied by a Hamming window function to reduce spectral leakage. Next, a 512-point short-time Fourier transform is performed on each frame to generate an amplitude spectrum, which is then fed into a lightweight convolutional neural network for abnormal voiceprint recognition.

[0030] In this implementation, the acquired vibration signal is decomposed using wavelet packet decomposition with five layers. The db4 wavelet basis function is selected, and the energy entropy of each sub-band is calculated as the feature vector. This vibration feature vector is concatenated with acoustic features (such as spectral centroid, zero-crossing rate, and Mel frequency cepstral coefficients) and input into a multimodal fusion classification model. This model adopts a dual-branch structure to process acoustic and vibration features separately, and finally outputs a comprehensive judgment result through a fully connected layer.

[0031] The analysis results are output in three ways: First, spectrograms and vibration time-domain waveforms are plotted in real time on the local display screen connected to the embedded control board 2, and the occurrence time of abnormal events is marked on the time axis; Second, operators can connect a USB flash drive or laptop through the data transmission interface 16 to export a structured test report, which includes metadata in JSON format (such as product model, test time, voltage setting value, and abnormality category) and spectrograms and waveforms in PNG format; Third, if the system determines that there is an abnormal sound pattern (such as confidence level exceeding 90%), it immediately triggers the built-in buzzer to emit an alarm sound for 1 second, and uploads the alarm log (including timestamp, feature vector summary, and judgment result) to the enterprise's remote server through the network port 15 for quality traceability.

[0032] Throughout the testing process, the power module 17 was set to output voltage of 12V DC via the voltage adjustment knob 11, simulating the motor's operation under rated conditions to ensure that the acoustic characteristics reflected the real-world usage scenario. After the test, the operator only needed to press the release button on the quick-clamp mechanism to remove the contour bracket 4 and replace it with a contour bracket suitable for the next type of product (such as a water pump or relay). There was no need to recalibrate the overall coordinate system or sensor parameters. Tooling switching and basic positioning could usually be completed within 3 minutes, enabling high-efficiency, multi-category continuous testing operations.

[0033] A method of using a portable AI acoustic spectrum analyzer device as described in claim 1 includes the following steps: S1. Tooling assembly: Take the contour bracket 4 out of the box body 1, align the positioning pin hole at its bottom with the positioning pin on the tooling base 3 and insert it, and lock it with the quick clamp mechanism; fix the micro vibration sensor 5 in the mounting hole reserved at the top of the contour bracket 4 with M3 screws; S2. Sound collector positioning: unfold the horizontal bracket 8 and rotate it around the hinge axis to a horizontal position, loosen the locking knob 10 of the universal ball joint 9, move the sound collector 6 to a distance of 5-20cm from the sound-emitting part of the product to be tested, and adjust its direction to face the sound-emitting surface to be tested, and then tighten the locking knob 10. S3. Cable connection: Insert one end of the power supply cable of the product under test into the power interface 12 of the corresponding specification, and connect the other end to the power input terminal of the product under test; insert the audio cable of the audio collector 6 into the audio interface 13 of the audio collector; insert the signal cable of the miniature vibration sensor 5 into the BNC interface 14 of the vibration sensor. S4. Start Acquisition: Start the main control program of the embedded control board 2, and start the high-fidelity audio acquisition channel and vibration signal acquisition channel; the sound collector 6 converts physical sound waves into analog electrical signals, which are then converted into 24-bit / 96kHz digital audio streams by the high-fidelity audio codec; the analog voltage signal output by the miniature vibration sensor 5 is conditioned by the isolation amplifier circuit and digitized by the ADC module of the embedded control board 2 at a sampling rate of 1MHz; S5. Signal Processing: The embedded control board 2 performs the following operations: (a) bandpass filtering on the digital audio stream, with the cutoff frequency set to 100Hz–10kHz; (b) background noise separation of the audio signal is performed using a deep noise reduction network based on the U-Net architecture; (c) Perform frame segmentation processing on the audio signal with a frame length of 25ms and a frame shift of 10ms, and apply a Hamming window to each frame; (d) Calculate the short-time Fourier transform of each frame, generate an amplitude spectrum, and input it into a pre-trained convolutional neural network for abnormal voiceprint recognition; (e) Simultaneously perform wavelet packet decomposition on the vibration signal, extract energy entropy features, fuse them with acoustic features, and input them into a multimodal classification model; S6, Result Output: The embedded control board 2 outputs the analysis results in the following ways: (i) Draw real-time spectrograms, vibration time-domain waveforms, and abnormal event markers on the local display screen; (ii) Export a structured detection report through the data transmission interface 16; (iii) If an abnormal voiceprint pattern is detected, trigger a buzzer alarm and upload an alarm log to a remote server through the network port 15.

[0034] The advantages of this invention are that this technical solution solves the problems of poor equipment portability, low tooling adaptation efficiency, weak noise suppression capability and low level of intelligent analysis in the acoustic testing of multiple product categories in industrial sites by integrating modular tooling, multi-degree-of-freedom sound acquisition structure, high-fidelity acquisition link and embedded AI inference engine. The tooling components use a quick-change contour bracket and a standardized base. For different products, only the contour bracket needs to be replaced, without the need to recalibrate the whole machine, which significantly shortens the on-site deployment time. The microphone bracket has a three-level adjustment structure of axial, lateral and universal, which enables the microphone to be precisely positioned in three-dimensional space, ensuring that it can still be aligned with the best pickup point even under complex product shapes, thereby improving the signal-to-noise ratio. The embedded control board's built-in deep noise reduction network and multimodal analysis model enable edge-side intelligent inference without the need for external server support, avoiding data upload delays and privacy leak risks.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A portable AI acoustic spectrum analyzer device, characterized in that: It includes a housing body, an embedded control board placed inside the housing body, a tooling assembly set on the side of the embedded control board, a power supply module for supplying power to the product under test and driving the whole machine to run, and a data acquisition and connection module for collecting sound and vibration signals. A sound collector bracket is set on one side of the tooling assembly. The embedded control board integrates an artificial intelligence-based acoustic spectrum analysis and processing system and a network communication module. The embedded control board is electrically connected to the power module and the acquisition and connection module via wiring harnesses. The acquisition and connection module includes a network port, a data transmission interface, a dedicated audio interface for the sound collector, and a dedicated BNC interface for the vibration sensor, all located on the front side panel of the enclosure.

2. The portable AI acoustic spectrum analyzer device according to claim 1, characterized in that: A voltage adjustment knob is provided on one side of the embedded control board, and the voltage adjustment knob is connected to the adjustable voltage regulator circuit in the power supply module through a wire.

3. The portable AI acoustic spectrum analyzer device according to claim 1, characterized in that: The power module includes a power supply body, a power support fixedly installed on the inner wall of the enclosure body, and four sets of standard power interfaces leading out from the power supply body and passing through the side wall of the enclosure body. Each set of power interfaces is connected to the corresponding power supply terminal of the product under test through a shielded wire harness.

4. The portable AI acoustic spectrum analyzer device according to claim 1, characterized in that: The network port is connected to the network controller of the embedded control board via a gigabit Ethernet PHY chip; the data transmission interface is connected to the data bus of the embedded control board via the controller; the dedicated audio interface for the sound collector is connected to the digital signal processor of the embedded control board via a high-fidelity audio codec; and the dedicated BNC interface for the vibration sensor is connected to the analog input channel of the embedded control board via an isolation amplifier circuit.

5. The portable AI acoustic spectrum analyzer device according to claim 1, characterized in that: The tooling assembly includes a tooling base fixedly mounted on the bottom plate of the housing body, a contouring bracket detachably snapped onto the upper surface of the tooling base, and a miniature vibration sensor fixed to the top of the contouring bracket by threaded fasteners; the contouring bracket is customized according to the shape of the product to be tested, and its bottom is provided with a positioning pin hole, which cooperates with the positioning pin provided on the tooling base.

6. The portable AI acoustic spectrum analyzer device according to claim 1, characterized in that: The microphone bracket includes an axial bracket vertically fixed to the bottom plate of the housing body, a transverse bracket hinged to the top of the axial bracket, and a microphone mounting base connected to the end of the transverse bracket via a universal ball joint; the microphone is screwed into the microphone mounting base by threads and connected to the microphone's dedicated audio interface via an audio cable; the universal ball joint includes a locking knob.

7. The portable AI acoustic spectrum analyzer device according to claim 5, characterized in that: The tooling base is equipped with a quick-clamp mechanism for locking the contour bracket onto the tooling base.

8. The method of using a portable AI acoustic spectrum analyzer device according to claims 1-7, characterized in that: Includes the following steps: S1. Tooling assembly: Take the contour bracket out of the box body, align the positioning pin hole at its bottom with the positioning pin on the tooling base, and lock it in place using the quick-clamp mechanism; fix the micro vibration sensor in the mounting hole reserved at the top of the contour bracket using M3 screws; S2. Sound collector positioning: unfold the horizontal bracket and rotate it around the hinge axis to a horizontal position, loosen the locking knob of the universal ball joint, move the sound collector to a distance of 5-20cm from the sound-emitting part of the product to be tested, and adjust its direction to face the sound-emitting surface to be tested, and then tighten the locking knob. S3. Cable connection: Insert one end of the power supply cable of the product under test into the power interface of the corresponding specification, and connect the other end to the power input terminal of the product under test; insert the audio cable of the sound collector into the dedicated audio interface of the sound collector; insert the signal cable of the miniature vibration sensor into the dedicated BNC interface of the vibration sensor. S4. Start Acquisition: The main control program of the embedded control board is started, and the high-fidelity audio acquisition channel and vibration signal acquisition channel are started; the sound collector converts physical sound waves into analog electrical signals, which are then converted into 24-bit / 96kHz digital audio streams by the high-fidelity audio codec; the analog voltage signal output by the miniature vibration sensor is conditioned by the isolation amplifier circuit and then digitized by the ADC module of the embedded control board at a sampling rate of 1MHz; S5. Signal Processing: The embedded control board performs the following operations: (a) bandpass filtering on the digital audio stream, with the cutoff frequency set to 100Hz–10kHz; (b) background noise separation of the audio signal using a deep noise reduction network based on the U-Net architecture; (c) framing the audio signal with a frame length of 25ms and a frame shift of 10ms, and applying a Hamming window to each frame; (d) calculating the short-time Fourier transform of each frame, generating an amplitude spectrum, and inputting it into a pre-trained convolutional neural network for abnormal voiceprint recognition. (e) Simultaneously perform wavelet packet decomposition on the vibration signal, extract energy entropy features, and input them into the multimodal classification model after fusing with acoustic features; S6. Result Output: The embedded control board outputs the analysis results in the following ways: (i) drawing real-time spectrograms, vibration time-domain waveforms and abnormal event markers on the local display screen; (ii) exporting a structured detection report through the data transmission interface; (iii) if an abnormal sound pattern is detected, triggering a buzzer alarm and uploading alarm logs to the remote server through the network port.

9. The method of using a portable AI acoustic spectrum analyzer device according to claim 8, characterized in that: In step S5, the training dataset of the deep noise reduction network includes more than 1,000 normal sound samples of products collected under different industrial noise environments, and is optimized using an adversarial training strategy.

Images