Metal material identification and young's modulus prediction device and method based on audible sound waves

Abstract

This invention belongs to the field of physics experiment teaching and testing technology, specifically a device and method for metal material identification and Young's modulus prediction based on audible sound waves. The prediction device includes: an excitation module for generating an acoustic excitation signal covering the target frequency band for the object under test; a sound acquisition module for acquiring the sound signal emitted by the collision of the metal object under test; and a data processing terminal for executing control of the excitation component and the sound acquisition module, and calling a trained metal acoustic feature classification model to complete the training and prediction tasks. This invention has a simple structure, low cost, and small size, making it easy to build and popularize, facilitating rapid deployment in resource-constrained teaching environments or confined industrial workstations. It also offers strong teaching intuitiveness: transforming abstract material properties into audible sound signals and visible recognition results, making it easier to stimulate student interest. This invention can be used as an AI physics experiment teaching device to establish AI+ education and teaching projects.

Description

Technical Field

[0001] This invention belongs to the field of physics experiment teaching and testing technology, specifically relating to a device and method for rapid identification of metallic materials and prediction of Young's modulus based on audible sound waves. Background Technology

[0002] For materials with similar appearances but unknown materials, determining their material composition is often linked to measuring their elastic properties, such as Young's modulus. Young's modulus is a key physical parameter characterizing a solid material's resistance to elastic deformation, and its accurate measurement is crucial in materials science research, engineering design, and quality control. In experimental teaching of physics at universities, measuring Young's modulus is also a classic experiment aimed at helping students understand the elastic behavior of materials.

[0003] Currently, traditional methods for accurately measuring Young's modulus mainly include static tensile method, bending method, and dynamic vibration method. [1] The static tensile method involves applying tension to a rod-shaped or filamentous sample and measuring its minute elongation using an optical lever method or displacement sensor to calculate Young's modulus. While this method is based on classic principles and yields reliable results, its experimental setup is typically large and complex (requiring precision force gauges and optical measurement systems), and it has specific shape requirements for the sample. The experimental process is also cumbersome and time-consuming, often requiring several class periods to complete. This limits its widespread application in general education or demonstration experiments for non-physics majors.

[0004] To achieve rapid and non-destructive testing of material elasticity, academia and industry have developed various modern testing techniques based on acoustic principles. Among them, ultrasonic testing is a widely used technique that calculates elastic parameters such as Young's modulus, shear modulus, and Poisson's ratio by measuring the longitudinal and transverse wave velocities of ultrasonic waves propagating in a material. This method has the advantages of being non-destructive and rapid, but it also has high requirements for the shape of the sample being measured. The measuring equipment, such as ultrasonic flaw detectors and high-frequency oscilloscopes, is relatively expensive, and operators need to receive certain professional training. In addition, to compensate for interfacial impedance mismatch, coupling agents may be needed when testing certain materials. Furthermore, the frequency of ultrasonic waves is far beyond the range of human hearing, making the entire measurement process less intuitive for beginners and less suitable for use as a general education or demonstration tool.

[0005] In recent years, with the rapid development of artificial intelligence technology, using machine learning algorithms to analyze the physical or chemical data of materials to predict their macroscopic properties has become a new research hotspot. Some studies have attempted to apply machine learning to the prediction of elastic modulus. For example, for high-entropy alloys (HEA) and complex-composition alloys (CCA), researchers have established mapping models between compositional characteristics and modulus using algorithms such as support vector regression (SVR), Gaussian process regression (GPR), gradient boosting, and polynomial regression (PR). Studies have shown that the second-order polynomial regression model exhibits extremely high accuracy in HEA prediction (R²=0.95), while the gradient boosting model has the best predictive ability when dealing with CCA. [2][3] Furthermore, in petroleum engineering, researchers have achieved high-precision inversion of the static Young's modulus of reservoir sandstone (R=0.96-0.999) using the Adaptive Neural Fuzzy Inference System (ANFIS) and the Mamdani Fuzzy Interferometric System (M-FIS). [4] These methods demonstrate the enormous potential of machine learning in the field of materials performance prediction, but most of them focus on starting from the microscopic composition of materials or theoretical calculation data, relying on complex input features. Their research purpose is mainly to design and screen new materials, rather than to build an AI physics experiment teaching instrument that can be used for physics experiment teaching and can intuitively demonstrate the entire process of "physical phenomena-data acquisition-intelligent analysis".

[0006] Furthermore, the industrial sector also has a high demand for distinguishing between different types of metals. For alloys with similar appearance, color, density, and other properties, it is difficult to quickly and non-destructively reclassify them when labels are lost. Currently, the mainstream rapid classification method on the market is spectral analysis. This method uses handheld metal element analyzers and other equipment to accurately measure the metal composition through laser irradiation. However, the measuring instruments used in this method are expensive, with individual instruments costing tens to hundreds of thousands of RMB, hindering widespread adoption.

[0007] In summary, for the tasks of rapid identification of metallic materials and prediction of Young's modulus, existing technologies for teaching Young's modulus include limitations such as complex equipment and long processing times. While modern ultrasonic testing methods are fast, they suffer from high equipment costs and a lack of intuitiveness. Furthermore, cutting-edge machine learning prediction methods primarily focus on theoretical research at the microscopic level. Therefore, there is a lack of a rapid, low-cost algorithm and supporting experimental equipment for classifying metallic materials and predicting Young's modulus with high experimental teaching value. Our invention aims to achieve these goals by combining audible sound waves with machine learning technology. Since there is a relatively clear correspondence between Young's modulus and material properties, this invention can also be used industrially to achieve low-cost and rapid classification of metallic parts.

[0008] References:

[0009] [1] Xie Mingyu, Li Faxin. Research progress on measurement methods of elastic modulus and internal friction of solids [J]. Progress in Mechanics, 2022, 52(1), 33.

[0010] [2]KHAKUREL H, TAUFIQUE MFN, ROY A, et al. Machine learning assisted prediction of the Young's modulus of compositionally complex alloys[J]. Scientific Reports, 2021, 11, 17311.

[0011] [3]RADHIKA N, SABARINATHAN M, RAGUNATH S, et al. Machine learning based prediction of Young's modulus of stainless steel coated with highentropy alloys[J]. Results in Materials, 2024, 23, 100607.

[0012] [4]ALAKBARI FS, MAHMOOD SM, BAMUMEN SS, et al. New and highlyaccurate static Young's modulus model using machine learning techniques[J]. ACS Omega, 2024, 9(38), 40687-40706. Summary of the Invention

[0013] The purpose of this invention is to provide a device and method for identifying metal materials and predicting Young's modulus based on audible sound waves, so as to solve the technical problems in the prior art, such as the bulky and expensive Young's modulus measurement equipment for metal materials, the difficulty in achieving rapid and non-destructive testing on site, and the high entry barriers of traditional industrial classification methods (such as spectral analysis) equipment.

[0014] The device for metal material identification and Young's modulus prediction based on audible sound waves provided by this invention includes: an excitation module, a sound acquisition module, and a data processing terminal; both the excitation module and the sound acquisition module are connected to the data processing terminal; wherein:

[0015] The excitation module includes a soundproof enclosure, a motor, motion excitation components, a motor drive circuit, a power supply, and a base; wherein:

[0016] The motor is mounted on the base; the motion excitation component is connected to the motor shaft; the motor is connected to the power supply through a motor drive circuit, the motor drive circuit controls the motor start-stop and speed, and the power supply supplies power to the entire excitation module; the soundproof cover covers the motor and the motion excitation component.

[0017] The motion excitation component includes a container and a metal object to be tested; the container is connected to a motor shaft and rotates with the motor; the metal object to be tested is placed inside the container; the metal object to be tested is driven by mechanical motion and subjected to stable and repeatable physical excitation, resulting in continuous and statistically stable mutual collisions, emitting sound, that is, generating an acoustic excitation signal covering the target frequency band.

[0018] The sound acquisition module is a measuring microphone, which is located inside a soundproof enclosure and placed near the container containing the metal object to be tested, and is used to collect the sound signal emitted by the collision of the metal object to be tested.

[0019] The measuring microphone transmits the collected sound signals to the data processing terminal;

[0020] The data processing terminal is connected to the motor drive circuit of the excitation module and the measurement microphone data of the sound acquisition module, respectively. It is used to control the operation of the motion excitation component, collect the measurement microphone data, and process it. Specifically, the data processing terminal controls the excitation module and the sound acquisition module to use the motor to control the motion excitation component to move in a specified manner, causing the metal object under test to collide and emit sound, and to complete model training, and to call the trained metal acoustic feature classification model to complete the prediction task.

[0021] Furthermore:

[0022] The container of the motion excitation component is a cylindrical rotating drum. The metal object to be tested is made into small balls. Multiple metal balls are placed in the drum. After the drum rotates, it causes the small balls to collide inside the drum and generate a stable sound signal.

[0023] The container of the motion excitation component is not limited to a cylindrical rotating drum; any mechanical structure capable of achieving the aforementioned physical collision effect is within the scope of protection of this invention. For example:

[0024] The motion excitation component can adopt a linear vibration structure: the excitation rod is driven by a motor to perform high-frequency reciprocating motion, which drives the metal object located in the constrained space to perform one-dimensional or multi-dimensional collision excitation.

[0025] The motion excitation component can adopt a spiral guide structure: a rotating hollow spiral tube drives the metal particles inside to move along a specific trajectory and generate continuous collisions.

[0026] In this invention, the feature classification model is trained using training audio files, and the trained model (including weights) is saved for performing material classification and Young's modulus prediction tasks on the predicted audio files.

[0027] Generally, before model training and inference, preprocessing algorithms (including but not limited to time-frequency transformation, noise reduction filtering, or envelope extraction) can be used on the original acoustic signal to transform non-stationary sound waves into feature vectors or feature spectra that the model can recognize.

[0028] The feature classification model can be based on machine learning algorithms, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs) and their variants; or other classification and regression models built based on supervised or semi-supervised learning. The feature classification model analyzes the extracted audio signal features, classifies the collected audio data according to metal material, and displays the material classification results and Young's modulus prediction results (including confidence index) on the user interface of the main program.

[0029] This invention also provides a method for metal material identification and Young's modulus prediction based on the above-mentioned prediction device, the specific steps of which are as follows:

[0030] (1) Activate the motion excitation component to generate collision sound; activate the sound acquisition module to receive sound signals;

[0031] (2) The data processing terminal receives and saves the digital audio data from the sound acquisition module as waveform files for training and prediction.

[0032] (3) Perform feature extraction on digital audio data (such as short-time Fourier transform, STFT);

[0033] (4) Input the extracted features into the metal acoustic feature classification model for training;

[0034] (5) Read digital audio data to perform prediction tasks;

[0035] (6) Displays the final metal material identification results.

[0036] In this invention, the metal object to be tested is a sphere, cylinder, or cube, etc.

[0037] The label corresponds to the material category of the metal object being tested. The label covers common metal materials in industrial and scientific research fields, including but not limited to: austenitic stainless steel (such as 304 stainless steel), martensitic stainless steel (such as 420 stainless steel), medium carbon steel (such as 45 steel), bearing steel, aluminum alloy (such as 1060 aluminum alloy), and copper alloys of various compositions (such as H59, H62, H65 brass), etc.

[0038] This invention aims to achieve the following technical objectives:

[0039] A novel acoustic fingerprint identification model and algorithm are constructed: using the sound generated by the collision of multiple metal objects as a signal, a machine learning model is trained through feature engineering techniques such as short-time Fourier transform (STFT) to establish a nonlinear mapping relationship between the macroscopic physical constants of metal materials (such as Young's modulus and density) and the acoustic characteristics of audible sound waves, thereby achieving high-precision classification of metal materials and prediction of Young's modulus.

[0040] Develop a low-cost, highly robust supporting measurement device: provide stable and repeatable physical excitation through a rotating mechanism, and achieve non-contact signal acquisition in conjunction with a high-sensitivity sound pickup module, reducing the stringent requirements on the environment and sample geometry, and improving the efficiency and universality of material detection.

[0041] Expanding experimental and demonstration teaching applications based on modern scientific research paradigms: Providing a technical carrier for physics experimental teaching that can intuitively demonstrate the entire process of "physical phenomenon collection - digital feature extraction - intelligent model reasoning", effectively solving the problem that traditional experimental models cannot be promoted in general education and popular science promotion due to the complexity of equipment and operation.

[0042] Providing an industrial-grade metal classification and traceability solution: The algorithm and device used in this invention are lightweight, fast-response, and highly transferable, and can be extended to practical application scenarios such as parts classification, online verification of alloy models, and quality control on industrial production lines, filling the gap in low-cost metal material identification technology in the industrial field.

[0043] The technical solution provided by this invention enables rapid estimation of the material classification and Young's modulus of common metallic materials with low equipment cost, empowering AI+physics experiment teaching and possessing industrial application value.

[0044] Technical effect

[0045] Compared with existing experimental instruments for measuring Young's modulus, such as tensile and bending methods, or spectroscopic analysis methods used to distinguish metal materials, this device has the following advantages:

[0046] Simple structure, low cost, and small size: Compared with traditional large instruments, this device can use a general-purpose stepper motor and a common measuring microphone, making it easy to build and popularize.

[0047] The device features a compact desktop design, which significantly reduces its footprint in laboratories and facilitates rapid deployment in resource-constrained teaching environments or small industrial workstations. Its non-contact data acquisition method reduces the device's sensitivity to vibration stability during installation, giving it excellent environmental adaptability.

[0048] The teaching is highly intuitive: it transforms the abstract "material properties" into audible "sound signals" and visible "recognition results", making it easier to stimulate students' interest.

[0049] Integrating cutting-edge technologies: Introducing AI technologies such as machine learning into physics experiments can serve as an AI physics experiment teaching device for launching AI+ education and teaching projects.

[0050] Short experimental time: In professional physics lab courses, Young's modulus experiments often require several class periods, which is not conducive to promoting the technology to non-physics majors. This method can compress the explanation of experimental principles and data measurement and processing into three class periods, making it easier to conduct experimental teaching for non-physics majors.

[0051] This invention is not only a teaching instrument, but its core algorithm and device also possess the technological potential for direct transfer to the industrial field: This invention utilizes low-cost acoustic sensors to accurately identify similar alloys (such as different series of stainless steel or brass), significantly reducing the entry barrier and operation and maintenance costs for industrial testing. In the future, a web-based remote software interface can be provided to enable online identification and classification of batches of metal components. Attached Figure Description

[0052] Figure 1 This is a simplified schematic diagram of the structure of the present invention.

[0053] Figure 2 This is a schematic diagram of the excitation module in an embodiment.

[0054] Figure 3 The image shown is a physical representation of an example.

[0055] Figure 4 This is a physical diagram of the excitation module in the embodiment.

[0056] Figure 5 The algorithm flowchart is shown in the example.

[0057] Figure 6 The audio waveforms and spectrum diagrams of small balls made of different materials are shown in the examples. Detailed Implementation

[0058] The present invention will be further described below through examples.

[0059] The present invention provides an algorithm and experimental device for predicting Young's modulus of materials based on audible sound waves and machine learning. The device includes a soundproof cover, a motor, a motion excitation component, a motor drive circuit, a power supply, a base, a sound acquisition module, and a data processing terminal.

[0060] The motor is mounted on the base via a fixed bracket; the motion excitation component includes a container and a metal object to be tested; the container is a rotating drum; the soundproof cover covers the motor and the motion excitation component.

[0061] A coupling is provided on the motor shaft;

[0062] The rotating drum body is a cylindrical body with one open end. A detachable end cap is provided at the open end. The end cap is connected by a limiting buckle structure or a foolproof structure to ensure that the end cap and the rotating drum body are fixed and will not rotate relative to each other. A regular hexagonal groove is provided at the center of the end cap, and a matching regular hexagonal protrusion (e.g., ...) is provided at the end of the coupling. Figure 4 As shown, the coupling end is detachably inserted into a regular hexagonal groove in the end cover. Strong magnets are provided inside the groove and at the coupling end. The rotating drum is strengthened by the attraction between the magnets on the end cover in the groove and the magnets at the coupling end. This allows the rotating drum to align with the motor shaft, which is supported by the motor's fixed bracket. A protruding shaft is located at the center of the other end of the rotating drum, and the shaft is mounted on a base via a drum bracket. The rotating drum rotates as the motor operates. The rotating drum can be removed and the end cover opened to replace the metal object being tested.

[0063] The rotating drum uses a composite connection method of "hexagonal groove + magnetic attraction" to solve the problem of radial runout caused by frequent start-stop or speed switching of the stepper motor. It not only enhances the axial stability of power transmission, but also provides a flexible connection effect similar to "automatic centering", which can offset the slight coaxiality deviation between the motor shaft and the drum axis, avoid stray noise caused by mechanical jamming from mixing into the acoustic sample, and ensure the purity of feature extraction.

[0064] The motor is connected to a power supply via a motor drive circuit. The motor drive circuit controls the motor's start-stop and speed, while the power supply provides power to the entire excitation module. Figure 2 As shown;

[0065] The metal object to be tested is a small metal ball placed inside a rotating drum. The metal object to be tested is driven by mechanical motion and subjected to stable and repeatable physical excitation, resulting in continuous and statistically stable mutual collisions, emitting sound, that is, generating an acoustic excitation signal covering the target frequency band.

[0066] The sound acquisition module is a measuring microphone, which is placed near the container containing the metal object to be tested, inside a soundproof enclosure, and is used to collect the sound signal emitted by the collision of the metal object to be tested.

[0067] The measuring microphone transmits the collected sound signals to the data processing terminal;

[0068] The data processing terminal is a computer, which is connected to the motor drive circuit and the measuring microphone via a USB data cable. The computer sends control commands to the motor drive circuit and receives the sound signals collected by the measuring microphone for further processing.

[0069] Specifically:

[0070] The computer runs on Windows 11 and has at least two USB ports. One port is used to connect a microphone via a data cable, and the other port is used to connect a stepper motor control module via a data cable. The computer internally contains a main control program, as well as excitation module control programs, sound acquisition programs, and machine learning programs that are called by the main program. The main program provides a user-friendly control and display interface, calls the excitation module control program to send motion control signals to the stepper motor during experiments, calls the sound acquisition program to receive audio signals collected by the microphone and store them as files, and calls the machine learning program to complete training, metal material classification, and Young's modulus prediction tasks.

[0071] The stepper motor control module, model A3A-DRV-01 stepper motor controller, requires a 7-12V DC power supply and is connected to a data cable via a USB Type-C interface.

[0072] The stepper motor, model 42BYGH24-401A, has a torque of 0.24W, a step angle of 1.8°, and a current of 1A. It is fixed to one side of the base with four Allen screws. Four ordinary wires are connected to the stepper motor control module, corresponding to the A+, A-, B-, and B+ terminals of the module.

[0073] The rotating roller is 3D printed from PLA. It is a cylindrical body open at one end, with a removable end cap at the open end. This end cap is mounted on a fixed support via a protruding pivot. The cylindrical body has a diameter of 6cm and a length of 12.35cm (excluding the end cap). After installing the end cap, the total length (excluding the protruding pivot) is 12.70cm. The pivot is 1.60cm long.

[0074] The base is 3D printed in one piece and made of PLA to ensure sufficient structural strength and ease of processing. The base consists of a flat rectangular base plate and a motor mounting bracket and a roller bracket fixed to it. The motor mounting bracket is perpendicular to the base plate and has a central through hole for accommodating the output shaft of the stepper motor. Four mounting screw holes are arranged around the through hole for securing the stepper motor. The roller bracket is positioned opposite the motor mounting bracket, and its top has a U-shaped notch to rotatably support the shaft of the rotating roller.

[0075] The soundproof cover is made of plexiglass, which is transparent and easy to observe, while also isolating external environmental noise from interfering with the acquisition of sound signals.

[0076] The microphone, model ABX13 USB Audio, is a 1-channel, 16-bit, 48000Hz format microphone. It has no audio enhancement function. The microphone is fixed in a position relatively close to the rotating drum, aligned with the drum's orientation. The microphone can capture sound signals emitted from the rotating drum and transmit the audio signals to the computer.

[0077] The metal spheres, 20mm in diameter, are available in various materials, such as 304 stainless steel, 420 stainless steel, 1060 aluminum, 45# carbon steel, and GCR15 bearing steel. Ten or more spheres of each material are used. See Table 1 for details.

[0078] Table 1: Young's modulus of currently available materials:

[0079] .

[0080] The main program is a graphical user interface (GUI) application built on the computer using the Python language and the tkinter graphics library. This application provides users with an integrated operating platform encompassing device control, data acquisition, model training, and prediction. The main program allows users to control motor rotation and sound acquisition parameters, and displays the acquired waveforms and training / prediction results via sliders, dropdown menus, and buttons on the graphical interface. Specific functions are implemented by the main program calling the excitation module control program, the sound acquisition program, and the machine learning program. The specific functions of the excitation module control program, the sound acquisition program, and the machine learning program are as follows.

[0081] The excitation module control program establishes a connection with the stepper motor via a serial port and sends a pre-packaged instruction string in a specific format (e.g., "V1.0\n" sets the rotation speed to 1.0 r / s) to the stepper motor control module, causing the stepper motor to rotate at the set speed. Setting the rotation speed to 0 stops the motor from rotating.

[0082] The sound acquisition program can use a microphone to collect audio data of a preset duration. When saving the waveform file to the local computer directory, it automatically constructs a standardized filename (e.g., L0_304 stainless steel_v=1.0_n=10_153000.wav) based on the sample category, motor speed, and optional number of balls and timestamps. This ensures the standardization and traceability of the collected training and test set data. The acquired .wav audio files are automatically saved to a preset folder structure organized by category.

[0083] The machine learning program can train a machine learning model and use the trained model to classify metal materials and predict Young's modulus.

[0084] During the training phase, the program can be configured to batch read acquired labeled audio files, call a preprocessing module (e.g., based on the librosa library) to slice and perform short-time Fourier transform (STFT) on the audio data to extract acoustic feature spectra. Subsequently, these feature spectra and their labels are fed into a predefined convolutional neural network (CNN) model for iterative training, and the trained model weights are saved as a file (e.g., model.pth). Key metrics such as loss and accuracy during training are printed in real-time in the main program's output window.

[0085] During the prediction phase, the program loads a newly acquired audio file of unknown category, performs the same preprocessing steps as during training, and then calls the saved model weight file for forward inference. The model output represents the probability distribution of different metal categories. Based on the distribution, the program determines the most likely metal category and outputs the final recognition result (e.g.,

[0086] “[GCR15_steel | Young's modulus ≈ 210 GPa] (84.51%) : L1_GCR15_steel_v=1.1_n=10_154629.wav”);

[0087] The results are displayed in the main program's output window. By mapping the identification results to a preset database of Young's modulus standard values ​​for metals, the predicted Young's modulus value can be output.

[0088] The working process is described in detail below. A typical experimental procedure is as follows:

[0089] 1. The operator places 10 small metal balls made of No. 45 carbon steel into the rotating drum;

[0090] 2. Open the main program on your computer, select the correct serial port and microphone device, and click the "Connect" button;

[0091] 3. On the interface, set the category label to "carbon_steel", the motor speed to "1.5" r / s, and the recording duration to "10" seconds;

[0092] 4. Click the "Send Command" button. The computer sends a speed command to the stepper motor control module. The stepper motor starts to drive the rotating drum at a speed of 1.5 r / s. The metal balls collide inside the drum and make a sound.

[0093] 5. Click the "Start Recording Current Material" button. The microphone will start collecting sound signals and transmit them to the computer in real time via the USB interface. The program will start a separate recording thread (_record_thread) to temporarily store the audio data stream in memory.

[0094] 6. After the 10-second recording time ends, the program will automatically stop recording and save the audio data in memory as a .wav file named according to the above naming rules;

[0095] 7. Repeat the above steps to collect data from various materials and rotation speeds to form a training set for the machine learning model. Specifically, for a small ball made of carbon steel, the roller speed is set between 0-3.0 r / s, with each increment at 0.2 r / s, and the sound signal is collected for 10 seconds. Then, the small ball is replaced with bearing steel and 1060 aluminum, and the above steps are repeated. The collected sound signals are used as the training set data.

[0096] 8. Call the model training function to complete the training and saving of the model;

[0097] 9. When performing the metal material classification and Young's modulus prediction tasks, place small balls of unknown material without setting classification labels, and repeat steps 2-6 to collect several audio segments. Specifically: take 10 small balls of unknown material, set the rotation speed of the roller to 1.0-2.0 r / s, and record a 10-second audio segment every 0.1 r / s. Simultaneously, set the roller rotation speed to 0, bring the phone close to microphone 5 and play a piece of music, start recording, and record for another 10 seconds. This music will serve as a control to verify the robustness of the model. Afterward, change the material of the small balls to other materials and repeat the above steps. Then call the prediction function, and the interface will display the identification result of the unknown small ball material within a few seconds.

[0098] Through the above implementation methods, this device can effectively collect acoustic fingerprints of different metals under controlled excitation. Taking three common metals—carbon steel, bearing steel, and 1060 aluminum alloy—as examples, after collecting signals and training for each metal in the above manner, the device is expected to achieve a classification accuracy of over 90% when identifying new predicted sound samples. The experimental results intuitively demonstrate the intrinsic correlation between the macroscopic physical properties of materials (such as Young's modulus and density) and their microscopic acoustic responses, proving the feasibility and superiority of this technical solution.

[0099] The aforementioned material list is only used to explain the classification capabilities of this invention, and not an exhaustive list of protected materials. Based on the 'acoustic fingerprint-physical property' mapping architecture constructed by this invention, those skilled in the art can automatically classify more categories of ferrous metals, non-ferrous metals, and their alloys without changing the algorithm framework of this invention by increasing the capacity of the training sample set or introducing more diverse metal sample data, and simultaneously predict the Young's modulus of the corresponding materials. This data-driven cross-material identification logic demonstrates the high scalability and algorithm universality of this invention in industrial application. For those skilled in the art, several improvements can be made without departing from the principles of this invention, and these improvements should also be considered within the scope of protection of this invention.

Claims

1. An audible sound wave-based metal material identification and Young's modulus prediction device, characterized by, include: The system comprises an excitation module, a sound acquisition module, and a data processing terminal; both the excitation module and the sound acquisition module are connected to the data processing terminal; wherein: The excitation module includes a soundproof enclosure, a motor, motion excitation components, a motor drive circuit, a power supply, and a base; wherein: The motor is mounted on the base; the motion excitation component is connected to the motor shaft; the motor is connected to the power supply through a motor drive circuit, the motor drive circuit controls the motor start-stop and speed, and the power supply supplies power to the entire excitation module; the soundproof cover covers the motor and the motion excitation component. The motion excitation component includes a container and a metal object to be tested; the container is connected to a motor shaft and rotates with the motor; the metal object to be tested is placed inside the container; the metal object to be tested is driven by mechanical motion and subjected to stable and repeatable physical excitation, resulting in continuous and statistically stable mutual collisions, emitting sound, that is, generating an acoustic excitation signal covering the target frequency band. The sound acquisition module is a measuring microphone, which is located inside a soundproof enclosure and placed near the container containing the metal object to be tested, and is used to collect the sound signal emitted by the collision of the metal object to be tested. The measuring microphone transmits the collected sound signals to the data processing terminal; The data processing terminal is connected to the motor drive circuit of the excitation module and the measurement microphone data of the sound acquisition module, respectively. It is used to control the operation of the motion excitation component, collect the measurement microphone data, and process it. Specifically, the data processing terminal controls the excitation module and the sound acquisition module to use the motor to control the motion excitation component to move in a specified manner, causing the metal object under test to collide and emit sound, and to complete model training, and to call the trained metal acoustic feature classification model to complete the prediction task.

2. The metal material identification and Young's modulus prediction device according to claim 1, characterized in that: The container of the motion excitation component is a cylindrical rotating drum. The metal object to be tested is made into small balls, and multiple metal balls are placed in the drum. After the drum rotates, it causes the small balls to collide inside the drum, generating a stable sound signal; or... The motion excitation component adopts a linear vibration structure: a motor drives an excitation rod to perform high-frequency reciprocating motion, causing a metal object located in the constrained space to undergo one-dimensional or multi-dimensional collision excitation; or... The motion excitation component adopts a spiral guide structure: it uses a rotating hollow spiral tube to drive the metal particles inside to move along a specific trajectory and generate continuous collisions.

3. The metal material identification and Young's modulus prediction device according to claim 1, characterized by The feature classification model is trained using training audio files, and the trained model is saved for performing material classification and Young's modulus prediction tasks on the predicted audio files.

4. The metal material identification and Young's modulus prediction device according to claim 3, characterized by Before model training and inference, preprocessing algorithms are used on the raw acoustic signals, including time-frequency transformation, noise reduction filtering, or envelope extraction, to transform non-stationary sound waves into feature vectors or feature spectra that the model can recognize.

5. The metal material identification and Young's modulus prediction device according to claim 4, characterized in that, The feature classification model uses machine learning algorithms, including convolutional neural networks (CNNs) or recurrent neural networks (RNNs) or their variants; or other classification and regression models built based on supervised or semi-supervised learning.

6. The metal material identification and Young's modulus prediction device according to claim 5, characterized by The feature classification model analyzes the extracted audio signal features, then classifies the collected audio data according to the metal material, and displays the material classification results and Young's modulus prediction results on the user interface of the main program.

7. The metal material identification and Young's modulus prediction device according to claim 5, characterized by The metal object to be tested is a sphere, cylinder, or cube.

8. The device of claim 1, wherein the device is configured to determine a Young's modulus of the metal material based on the at least one of the first and second signals. The motor is mounted on the base via a fixed bracket; the container is a rotating drum; a coupling is provided on the motor shaft; The rotating drum body is a cylindrical tube open at one end, with a detachable end cap at the open end. The end cap is connected by a limiting buckle structure or a foolproof structure to ensure that the end cap and the rotating drum body are fixed and will not rotate relative to each other. The end cap has a regular hexagonal groove at its center, and a matching regular hexagonal protrusion is provided at the end of the coupling. The end of the coupling is detachably inserted into the regular hexagonal groove of the end cap. Magnets are provided inside the groove and at the end of the coupling. The rotating drum is strengthened by the attraction between the magnet at the groove of the end cap and the magnet at the end of the coupling. This allows the rotating drum to align with the motor shaft and be supported by the motor's fixed bracket. The other end of the rotating drum has a protruding shaft at its center, which is mounted on a base by a drum bracket. The rotating drum rotates as the motor operates.

9. The method of metal material identification and Young's modulus prediction based on the prediction device according to any one of claims 1 to 8, characterized in that, The specific steps are as follows: (1) Activate the motion excitation component to generate collision sound; activate the sound acquisition module to receive sound signals; (2) The data processing terminal receives and saves the digital audio data from the sound acquisition module as waveform files for training and prediction. (3) Extract features from digital audio data; (4) Input the extracted features into the metal acoustic feature classification model for training; (5) Read digital audio data to perform prediction tasks; (6) Displays the final metal material identification results.

Images