Detection method of wood structure health state detection device based on combination of acoustic features and deep learning

Abstract

The invention provides a detection method of a wood structure health state detection device based on combination of acoustic features and deep learning. The detection method comprises the following steps: establishing communication among units after a power supply is switched on, and presetting parameters; the electromagnetic automatic knocker knocks the wood structure test piece to be detected; the high-sensitivity microphone array collects sound signals generated by knocking the wood structure to be detected in real time, and the optical sensor measures the moving distance of the detection device; the edge calculation AI chip preprocesses the sound signal, operates an artificial intelligence algorithm and outputs a health state detection result of the wood structure to be detected, and the WIFI antenna transmits collected data and the detection result to a cloud end and a mobile phone end; and after the detection is completed, drawing a contour map by combining the recorded movement distance of the detection device and the detection results of different parts, and visually displaying the detection results. By knocking the surface of the wood structure and collecting the sound signal, the type and position of the internal defect of the wood structure can be automatically identified in combination with a deep learning algorithm.

Description

Technical Field

[0001] This invention relates to the technical field of devices and methods for detecting the health status of timber structures, and particularly to a device and method for detecting the health status of timber structures that combines acoustic features with deep learning. Background Technology

[0002] Currently, my country has entered a stage of high-quality development, but the safety situation of wooden structure cultural relics remains severe, and the technological innovation capabilities and application levels of wooden structure cultural relics urgently need to be improved. my country has a huge stock of wooden structures that urgently require maintenance and protection. Ancient Chinese architecture was primarily based on wood, but due to its biomass properties, it is prone to insufficient resistance and even collapse under the long-term effects of environmental erosion and loads. The preciousness of ancient buildings lies in their role as witnesses to the development of Chinese civilization. Through them, one can explore the civilization and cultural context hidden throughout the ages. At the same time, ancient buildings are also an important medium for attracting tourists from all over the world to understand traditional Chinese culture and history. Therefore, conducting research on wooden structure testing technology is beneficial not only to the protection of historical relics but also to the sustainable development of tourism and cultural undertakings in various regions.

[0003] Currently, there are relatively few methods for detecting internal defects in timber structures, mainly including visual inspection, micro-drilling resistance testing, and stress wave-based methods. Visual inspection relies heavily on operator experience and is highly subjective. Micro-drilling resistance testing is a common method for assessing the internal condition of timber components; however, it is semi-destructive, and the measurement results may be affected by the height of the test location. Stress wave-based methods typically use wave velocity as an indicator to obtain internal information, as the presence of internal pores causes a rapid decrease in wave velocity. However, this method requires the installation of sensors before each local test, making the process cumbersome, and the equipment is expensive and inconvenient to carry, greatly limiting its application in practical projects. Therefore, there is an urgent need to develop an efficient non-destructive testing technology to address the problems of destructiveness, inefficiency, and high cost associated with existing testing methods.

[0004] The percussion method is a non-destructive testing technique. Because it eliminates the need for sensor coupling and installation during data acquisition, it saves significant testing time and costs, and has been widely applied in fields such as railways, ceramics, and agricultural products. Developing a percussion-based testing technology for timber structures would enrich the theoretical and technical system of non-destructive testing for timber structures. Patent No. 202310584696.6 discloses a non-destructive testing device, system, and method for the interior of ancient wooden structures. This method obtains percussion force and sound signals by percussion on the surface of the timber structure, analyzing the presence or absence of voids and defects. However, this method primarily relies on manual feature extraction and defect identification, which leads to problems such as high data processing costs, high error rates, and low levels of intelligence when dealing with large datasets. Summary of the Invention

[0005] This invention provides a detection method for a wood structure health status detection device based on the combination of acoustic features and deep learning. It can automatically identify the type and location of internal defects in the wood structure by tapping the surface of the wood structure and collecting sound signals, combined with deep learning algorithms.

[0006] The above-mentioned objectives of the present invention are achieved through the following technical solutions:

[0007] This invention provides a detection method for a timber structure health status detection device based on the combination of acoustic features and deep learning. The timber structure health status detection device includes a striking and sound acquisition unit, an optical ranging unit, a control and processing unit, and a human-computer interaction unit. The striking and sound acquisition unit includes an electromagnetic automatic striking device, an adjustable bracket, adjustable pulleys, and a high-sensitivity microphone array. The optical ranging unit is an optical sensor. The control and processing unit includes an edge computing AI chip, a microcontroller, and a battery. The human-computer interaction unit includes a touchscreen and a WIFI antenna. The adjustable pulley is formed by connecting pulley modules via ball joints. The pulley modules are connected to a cylindrical connecting rod, and the adjustable pulleys are installed below the adjustable bracket. The high-sensitivity microphone array is formed by connecting microphone modules via ball joints. The microphone modules are connected to a cylindrical connecting rod, and the high-sensitivity microphone array is installed above the adjustable bracket. One end of the adjustable bracket is connected to the electromagnetic automatic striking device. The other end is configured as a handle; the electromagnetic automatic tapper includes a housing frame I, a motor, a pressure ceramic sensor, and an adjustable tapping hammer; the motor and pressure ceramic sensor are installed inside the housing frame I, the motor is installed on the upper part of the adjustable tapping hammer, and the motor is connected to the hammer via a drive shaft; the pressure ceramic sensor is installed on the upper part of the adjustable tapping hammer and is connected to the housing frame I; the adjustable tapping hammer is connected to the housing frame I; the optical sensor is installed inside the housing frame I; the edge computing AI chip, microcontroller, and battery are all installed inside the handle of the adjustable bracket; the microcontroller is connected to the electromagnetic automatic tapper, high-sensitivity microphone array, optical sensor, and edge computing AI chip via a serial peripheral interface; the battery is connected to the electromagnetic automatic tapper, high-sensitivity microphone array, optical sensor, edge computing AI chip, touch screen, and WIFI antenna via a power cable; the touch screen is installed outside the adjustable bracket and is connected to the microcontroller via a serial port; the WIFI antenna is installed on the touch screen;

[0008] The detection method for a timber structure health status detection device based on acoustic features and deep learning includes the following steps:

[0009] Step 1: Connect the power supply, establish communication between each unit, and preset the tapping force and frequency parameters, as well as the sound acquisition parameters.

[0010] Step 2: Adjust the curvature of the high-sensitivity microphone array and adjustable pulley in the testing device according to the shape and size of the wooden structure. Then, place the testing device tightly against the wooden structure specimen to be tested, and use the microcontroller to control the electromagnetic automatic tapper to tap the wooden structure specimen to be tested according to the tapping force and frequency parameters preset in Step 1.

[0011] Step 3: The microcontroller controls the high-sensitivity microphone array to collect the sound signals generated by striking the wooden structure under test in real time according to the sound acquisition parameters preset in Step 1, and controls the optical sensor to measure the distance the detection device moves.

[0012] Step 4: The microcontroller controls the edge computing AI chip to preprocess the sound signal and run artificial intelligence algorithms, and outputs the health status detection results of the wooden structure to be detected. The WIFI antenna is controlled to transmit the collected data and detection results to the cloud and mobile phone.

[0013] Step 5: After data collection at a certain point is completed, slide the detection device vertically to another point using the adjustable pulley, and repeat steps 3 and 4.

[0014] Step Six: After completing the overall inspection of the wooden structure to be inspected, draw a contour map based on the recorded movement distance of the inspection device and the inspection results of different parts, and visualize the inspection results.

[0015] Furthermore, in step one, the striking force needs to be determined based on the hardness of the wooden structure to be tested. The striking force needs to be adjusted to be able to excite local vibrations of the structure without damaging its overall integrity. The striking force frequency parameter is set between 40Hz and 60Hz to achieve a balance between striking efficiency and the vibration duration of the wooden structure to be tested. The sound acquisition parameters include sampling frequency, frame length, and windowing method. The sampling frequency is set to be no less than twice the natural frequency of the wooden structure to be tested, the frame length is set between 512 and 2048, and the windowing method is preferably a Hanning window.

[0016] Furthermore, in step two, the curvature of the high-sensitivity microphone array and the adjustable pulley needs to be adjusted to be parallel to the diameter of the wooden structure to be tested.

[0017] Furthermore, in step three, the acquisition start time of the high-sensitivity microphone array is the time when the electromagnetic automatic tapper just begins to reach a tapping force of 0.5N when it taps the wooden structure under test, and the acquisition duration is 0.01~0.02s, thereby avoiding the acquisition of unnecessary signals.

[0018] Furthermore, in step four, the preprocessing operation of the sound signal includes noise reduction and feature extraction; the noise reduction is preferably performed using wavelet decomposition; the features are preferably short-time Fourier transform image features.

[0019] Furthermore, in step four, the artificial intelligence algorithm is one of MobileNetV3, ShuffleNetV2, EfficientNet, GhostNet, and SqueezeNet lightweight convolutional neural networks to ensure efficient operation under limited computing resources; the artificial intelligence algorithm is improved by introducing transfer learning technology and attention mechanism, thereby enhancing the algorithm's transfer learning ability and detection accuracy.

[0020] Furthermore, the transfer learning technique involves first pre-training the algorithm on the large-scale general dataset Wikipedia, and then using the pre-trained algorithm to train and fine-tune the algorithm's parameters on the timber structure detection data; the attention mechanism is the Squeeze-and-Excitation (SE) module.

[0021] Furthermore, in step four, the artificial intelligence algorithm needs to be trained indoors beforehand. The training dataset should take into account different variables such as defect type, defect size, defect location, striking force, striking location, structural form, structural size, and structural stress to ensure the generalization of the artificial intelligence algorithm.

[0022] Furthermore, in step four, the health status test result index of the wooden structure under test is the damage probability.

[0023] Furthermore, in step six, the tool for drawing contour maps is selected from Oringin, Matalab, or Python software; the color scheme of the contour maps should gradually change from blue to red as different damage probability intervals increase. The higher the damage probability value, the redder the color, indicating that the area is more likely to be damaged, thereby obtaining the health status of different areas of the wooden structure.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] (1) The high-sensitivity microphone array and adjustable pulley of the device of the present invention can adapt to different forms and sizes of the wooden structure to be tested, and can replace the manual knocking process to realize the automation of knocking and data acquisition. At the same time, through the telescopic function of the bracket, the influence of the wooden structure to be tested being too high or too low on the measurement process is reduced.

[0026] (2) The present invention provides a method for detecting wooden structures by tapping sound. Compared with visual inspection, the present invention can get rid of the influence of human subjectivity. Compared with micro-drill resistance method, the present invention will not damage the integrity of the wooden structure under test. Compared with stress wave based method, the present invention does not require the coupling installation and disassembly process of sensor, and the detection efficiency is significantly improved.

[0027] (3) The present invention uses a lightweight convolutional neural network, which maintains high accuracy while having the advantages of low computational cost and few parameters. Compared with traditional algorithms, it can better adapt to resource-constrained mobile devices. In addition, by introducing transfer learning technology and attention mechanism, the transfer learning ability and detection accuracy of the algorithm can be greatly improved.

[0028] (4) This invention provides the idea of ​​a high-sensitivity microphone array. Compared with the traditional single-point tapping and single-point acquisition method, it can realize efficient scanning tapping and data acquisition. In addition, by combining displacement measurement data, it can draw contour maps of damage probability on different detection surfaces, thereby realizing damage location and visualization.

[0029] (5) The present invention provides a mobile timber structure inspection device method, which has multiple functions such as automatic tapping, intelligent inspection, human-computer interaction, and remote viewing. The equipment has a simple structure, low cost, and is suitable for widespread application, which helps the intelligent transformation and upgrading of the construction field. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the detection device of the present invention.

[0031] Figure 2 for Figure 1 A cross-sectional schematic diagram of the electromagnetic automatic percussion device structure.

[0032] Figure 3 for Figure 1 A cross-sectional schematic diagram of the handle structure of the adjustable bracket.

[0033] Figure 4 for Figure 1 The different unit components form the framework diagram.

[0034] Figure 5 for Figure 1 The working principle framework diagram.

[0035] Figure 6 This is a visual structural diagram of the detection process using the present invention.

[0036] Explanation of markings in the diagram:

[0037] 1-Electromagnetic automatic hammer, 2-Adjustable stand, 3-Pulley, 4-Cylindrical connecting rod, 5-Touch screen, 6-WIFI antenna, 7-Work indicator light, 8-Switch button, 9-Data interface, 10-Microphone, 11-Optical sensor, 12-Motor, 13-Pressure ceramic sensor, 14-Adjustable hammer head, 15-1 Outer shell frame I, 15-2 Handle outer shell frame II, 16-Edge computing AI chip, 17-Microcontroller, 18-Battery, 19-Handle with adjustable stand. Detailed Implementation

[0038] The present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. It should be noted that the specific embodiments are not intended to limit the scope of the present invention.

[0039] like Figure 1-4 As shown, the present invention discloses a wood structure health status detection device combining acoustic features and deep learning, comprising a tapping and sound acquisition unit, an optical ranging unit, a control and processing unit, and a human-computer interaction unit; the tapping and sound acquisition unit includes an electromagnetic automatic tapper 1, an adjustable bracket 2, adjustable pulleys, and a high-sensitivity microphone array; the optical ranging unit is an optical sensor 11; the control and processing unit includes an edge computing AI chip 16, a microcontroller 17, and a battery 18; and the human-computer interaction unit includes a touch screen 5 and a WIFI antenna 6.

[0040] The adjustable pulley is formed by connecting pulley modules via ball joints. Each pulley module consists of a pulley 3 connected to a cylindrical connecting rod 4. The adjustable pulley is installed below the adjustable bracket. The adjustable pulley is used to move the entire testing device. The pulley modules are connected by ball joints, allowing for adjustment of the appropriate sliding radius according to different wooden structure shapes and sizes.

[0041] The high-sensitivity microphone array is composed of microphone modules connected by ball joints. Each microphone module consists of a microphone 10 connected to a cylindrical connecting rod 4. The high-sensitivity microphone array is mounted on an adjustable bracket. Different microphone modules are connected by ball joints to allow for adjustment according to the shape and size of the wooden structure. The high-sensitivity microphone array is used to collect vibration sound signals from different directions after the structure is struck.

[0042] The adjustable bracket 2 is connected to the electromagnetic automatic tapper 1 at one end, and the other end of the adjustable bracket is set as a handle 19. The adjustable bracket 2 is used to support the entire detection device. The electromagnetic automatic tapper 1 is located at one end of the adjustable bracket 2 and can freely adjust the tapping force and frequency for tapping the surface of the object being tested. The electromagnetic automatic hammer 1 includes a housing frame I 15-1, a motor 12, a pressure ceramic sensor 13, and an adjustable hammer head 14. The motor 12 and the pressure ceramic sensor 13 are installed inside the housing frame I 15-1. The motor 12 is installed on the upper part of the adjustable hammer head 14 and is connected to the hammer head 14 via a drive shaft. The pressure ceramic sensor 13 is installed on the upper part of the adjustable hammer head 14, and the edge surface of the pressure ceramic sensor 13 is connected to the housing frame I 15-1 by welding. The adjustable hammer head 14 is threaded onto the housing frame I 15-1. The motor 12 provides power to the adjustable hammer head 14, the pressure ceramic sensor 13 collects the force of the hammering process, and the adjustable hammer head 14 is used to strike the wooden structure to be tested. As a further preferred embodiment, the adjustable hammer head 14 is tightened and fixed to the outer shell frame I15-1 by rotating through the threaded hole, and is used to strike the wooden structure to be tested. The material can be plastic, rubber, steel or aluminum alloy, so that it can be appropriately selected according to the stiffness of different wooden structure materials to be tested, so that the stiffness of the wooden structure to be tested is close to that of the hammer head material.

[0043] As a further preferred embodiment, the adjustable bracket 2 has a Z-shaped appearance with a retractable structure in the middle, which can adjust the length of the bracket to meet the needs of different measurement heights.

[0044] As a further preferred embodiment, the adjustable pulley consists of at least three pulley modules connected by ball joints, thereby allowing for adjustment of the appropriate sliding arc according to different wooden structure shapes and sizes.

[0045] As a further preferred embodiment, the high-sensitivity microphone array includes at least five microphone modules, which are connected by ball joints to allow for adjustment according to the shape and size of the wooden structure, and are used to collect vibration sound signals from different directions after the structure is struck.

[0046] The optical sensor 11 is installed inside the outer frame I15-1 of the electromagnetic automatic tapper 1 and is used to record the distance the detection device moves.

[0047] The edge computing AI chip 16 is installed inside the handle 19 of the adjustable bracket 2. It can be selected from Google Corel and others to execute artificial intelligence algorithms, realize automatic processing of sound signals of wooden structures and intelligent classification of their health status.

[0048] The microcontroller 17 is installed inside the handle 19 of the adjustable bracket 2, and can be a Raspberry Pi or similar device. The microcontroller 17 is connected to the electromagnetic automatic tapper 1, the high-sensitivity microphone array, the optical sensor 11, and the edge computing AI chip 16 via a serial peripheral interface. The microcontroller 17 is used to control the electromagnetic automatic tapper 1, the high-sensitivity microphone array, the optical sensor 11, and the edge computing AI chip 16.

[0049] The battery 18 is installed in the handle 19 of the adjustable bracket 2. The battery 18 is connected to the electromagnetic automatic tapper 1, the high-sensitivity microphone array, the optical sensor 11, the edge computing AI chip 16, the touch screen 5 and the WIFI antenna 6 respectively via power cords, and is used to provide power to the electromagnetic automatic tapper 1, the high-sensitivity microphone array, the optical sensor 11, the edge computing AI chip 16, the touch screen 5 and the WIFI antenna 6.

[0050] The touchscreen 5 is mounted on the outside of the adjustable bracket 2, and the touchscreen 3 is connected to the microcontroller 17 via a serial port. The touchscreen 5 is equipped with a working indicator light 7, a power button 8, and a data interface 9, which are used for setting parameters for different modules and sending start / stop commands.

[0051] The WIFI antenna 6 is installed on the touch screen 5 and is used to transmit data to the cloud, mobile phone, etc., so as to facilitate remote viewing of data.

[0052] This specific embodiment also provides a detection method for a timber structure health status detection device based on the combination of acoustic features and deep learning, including the following steps:

[0053] Step 1: Connect the power supply, establish communication between each unit, and preset the striking force and frequency parameters, as well as the sound acquisition parameters. The striking force needs to be determined based on the hardness of the wooden structure under test, and the striking force needs to be adjusted to excite local vibrations in the structure without damaging its overall integrity. The striking force frequency parameter is set between 40Hz and 60Hz to achieve a balance between striking efficiency and the vibration duration of the wooden structure under test. The sound acquisition parameters include sampling frequency, frame length, and windowing method. The sampling frequency is set to be no less than twice the natural frequency of the wooden structure under test, the frame length is set between 512 and 2048, and the preferred windowing method is the Hanning window.

[0054] Step two: Adjust the curvature of the high-sensitivity microphone array and adjustable pulleys in the testing device according to the shape and size of the wooden structure. Then, place the testing device firmly against the wooden structure specimen to be tested. The microcontroller controls the electromagnetic automatic tapper to tap the specimen according to the tapping force and frequency parameters preset in Step one. The curvature of the high-sensitivity microphone array and adjustable pulleys needs to be adjusted to be parallel to the diameter of the wooden structure to be tested.

[0055] Step three: The microcontroller controls a high-sensitivity microphone array to collect sound signals generated by striking the wooden structure under test in real time according to the sound acquisition parameters preset in step one, and controls the optical sensor to measure the distance the detection device moves. The acquisition start time of the high-sensitivity microphone array is the moment when the electromagnetic automatic tapper just reaches a striking force of 0.5N when striking the wooden structure under test, and the acquisition duration is 0.01~0.02s, thereby avoiding the acquisition of unnecessary signals.

[0056] Step four: The microcontroller controls the edge computing AI chip to preprocess the sound signal and run artificial intelligence algorithms, and outputs the health status detection results of the wooden structure to be tested. The WIFI antenna is controlled to transmit the collected data and detection results to the cloud and mobile phone, etc. The health status detection result index of the wooden structure to be tested is the damage probability.

[0057] The preprocessing operations of the sound signal include noise reduction and feature extraction. The preferred noise reduction method is wavelet decomposition, and the feature extraction is short-time Fourier transform image features.

[0058] The wavelet decomposition method mentioned above is derived from: Liu Huiyong, Zhang Song, Li Jianfeng, et al. Tool wear condition monitoring using an improved CNN-BiLSTM model [J]. Chinese Mechanical Engineering, 2022, 33(16):1940-1947+1956.

[0059] The calculation process for the short-time Fourier transform image features is as follows: For a non-stationary signal First, construct a time window function. Then through observe To obtain a local signal over a certain time period When the time window is short enough, This can be viewed as a locally stationary signal. Next, we will... by Continue translating along the timeline with the center in mind to obtain multiple For all Perform a Fourier transform to obtain The function after STFT transformation:

[0060]

[0061] In the formula, For window functions; For the frequency of time-domain analysis; This represents the time for time-domain analysis.

[0062] bandwidth It can be calculated using the following formula:

[0063]

[0064] When the frequency spacing between two sine waves is greater than At that time, these two sine waves can be distinguished, therefore, This is known as the frequency resolution of the STFT. Similarly, the resolution in the time domain... for:

[0065]

[0066] In the formula, the denominator is Energy.

[0067] Similar to the principle of frequency resolution, when the time interval between two pulse signals is greater than... Then, these two pulse signals are distinguished. The time resolution is referred to as STFT. However, and The two cannot be arbitrarily small at the same time; according to the Heisenberg inequality, they are subject to the following conditions:

[0068]

[0069] The equation holds if and only if the window function is a Gaussian function.

[0070] The artificial intelligence algorithm is one of the lightweight convolutional neural networks such as MobileNetV3, ShuffleNetV2, EfficientNet, GhostNet, and SqueezeNet, to ensure efficient operation under limited computing resources. The artificial intelligence algorithm is improved by introducing transfer learning technology and attention mechanism, thereby enhancing the transfer learning ability and detection accuracy of the algorithm.

[0071] The transfer learning technique involves first pre-training the algorithm on the large-scale general dataset Wikipedia, and then using the pre-trained algorithm to train and fine-tune the algorithm's parameters on the timber structure detection data. The attention mechanism is the Squeeze-and-Excitation (SE) module.

[0072] The Squeeze-and-Excitation (SE) module is derived from: Li Jingyu, Feng Zhongxiang, Zhang Weihua, et al. A method for constructing driver information cognitive maps under the background of intelligent connected vehicles [J]. China Journal of Highway and Transport, 2023, 36(09):302-314.

[0073] The artificial intelligence algorithm needs to be trained indoors in advance. The training dataset should take into account different variables such as defect type, defect size, defect location, striking force, striking location, structural form, structural size, and structural stress to ensure the generalization of the artificial intelligence algorithm.

[0074] Step 5: After data collection at a certain point is completed, slide the detection device vertically to another point using the adjustable pulley, and repeat steps 3 and 4.

[0075] Step Six: After completing the overall inspection of the wooden structure to be inspected, draw a contour map based on the recorded movement distance of the inspection device and the inspection results of different parts, and visualize the inspection results.

[0076] like Figure 6 As shown, contour maps of damage probability were plotted using Oringinr software based on the detection results of different locations. The color scheme of the contour maps gradually changes from blue to red as the damage probability range increases. It can be observed that... Figure 6 The lower middle section shows a distinct red area, indicating a high probability of damage in that area, while the other areas are closer to blue, suggesting a lower probability of damage. These results effectively reflect the health status of different areas of the wooden structure.

Claims

1. A detection method for a timber structure health status detection device based on acoustic features and deep learning, characterized in that: The acoustic feature and deep learning-integrated timber structure health status detection device includes a striking and sound acquisition unit, an optical ranging unit, a control and processing unit, and a human-computer interaction unit. The striking and sound acquisition unit includes an electromagnetic automatic striking device, an adjustable bracket, adjustable pulleys, and a high-sensitivity microphone array. The optical ranging unit is an optical sensor. The control and processing unit includes an edge computing AI chip, a microcontroller, and a battery. The human-computer interaction unit includes a touchscreen and a Wi-Fi antenna. The adjustable pulley is formed by pulley modules connected by ball joints, and each pulley module is connected to a cylindrical connecting rod. The adjustable pulley is installed below the adjustable bracket. The high-sensitivity microphone array is formed by microphone modules connected by ball joints, and each microphone module is connected to a cylindrical connecting rod. The high-sensitivity microphone array is installed above the adjustable bracket. One end of the adjustable bracket is connected to the electromagnetic automatic striking device, and the other end is configured as a handle. The electromagnetic automatic striking device... The device includes a housing frame I, a motor, a pressure ceramic sensor, and an adjustable hammerhead. The motor and pressure ceramic sensor are mounted inside the housing frame I. The motor is mounted on top of the adjustable hammerhead and connected to the hammerhead via a drive shaft. The pressure ceramic sensor is mounted on top of the adjustable hammerhead and connected to the housing frame I. The adjustable hammerhead is connected to the housing frame I. An optical sensor is mounted inside the housing frame I. An edge computing AI chip, a microcontroller, and a battery are all mounted inside the handle of the adjustable stand. The microcontroller is connected to the electromagnetic automatic tapper, a high-sensitivity microphone array, the optical sensor, and the edge computing AI chip via a serial peripheral interface. The battery is connected to the electromagnetic automatic tapper, the high-sensitivity microphone array, the optical sensor, the edge computing AI chip, a touchscreen, and a WIFI antenna via a power cable. The touchscreen is mounted outside the adjustable stand and is connected to the microcontroller via a serial port. The WIFI antenna is mounted on the touchscreen. The detection method for a timber structure health status detection device based on acoustic features and deep learning includes the following steps: Step 1: Connect the power supply, establish communication between each unit, and preset the tapping force and frequency parameters, as well as the sound acquisition parameters. Step 2: Adjust the curvature of the high-sensitivity microphone array and adjustable pulley in the testing device according to the shape and size of the wooden structure. Then, place the testing device tightly against the wooden structure specimen to be tested, and use the microcontroller to control the electromagnetic automatic tapper to tap the wooden structure specimen to be tested according to the tapping force and frequency parameters preset in Step 1. Step 3: The microcontroller controls the high-sensitivity microphone array to collect the sound signals generated by striking the wooden structure under test in real time according to the sound acquisition parameters preset in Step 1, and controls the optical sensor to measure the distance the detection device moves. Step 4: The microcontroller controls the edge computing AI chip to preprocess the sound signal and run artificial intelligence algorithms, and outputs the health status detection results of the wooden structure to be detected. The WIFI antenna is controlled to transmit the collected data and detection results to the cloud and mobile phone. Step 5: After data collection at a certain point is completed, slide the detection device vertically to another point using the adjustable pulley, and repeat steps 3 and 4. Step Six: After completing the overall inspection of the wooden structure to be inspected, draw a contour map based on the recorded movement distance of the inspection device and the inspection results of different parts, and visualize the inspection results.

2. The detection method of the timber structure health status detection device based on acoustic features and deep learning according to claim 1, characterized in that: In step one, the striking force needs to be determined based on the hardness of the wooden structure to be tested. The striking force needs to be adjusted to be able to excite local vibrations of the structure without damaging its overall integrity. The striking force frequency parameter is set between 40Hz and 60Hz to achieve a balance between striking efficiency and the vibration duration of the wooden structure to be tested. The sound acquisition parameters include sampling frequency, frame length, and windowing method. The sampling frequency is set to be no less than twice the natural frequency of the wooden structure to be tested, the frame length is set between 512 and 2048, and the windowing method is preferably a Hanning window.

3. The detection method of the timber structure health status detection device based on acoustic features and deep learning according to claim 1, characterized in that: In step two, the curvature of the high-sensitivity microphone array and the adjustable pulley needs to be adjusted to be parallel to the diameter of the wooden structure to be tested.

4. The detection method of the timber structure health status detection device based on acoustic features and deep learning according to claim 1, characterized in that: In step three, the high-sensitivity microphone array starts collecting data at the moment when the electromagnetic automatic tapper strikes the wooden structure under test and the striking force just reaches 0.5N. The collection time is 0.01~0.02s, thereby avoiding the collection of unnecessary signals.

5. The detection method of the timber structure health status detection device based on acoustic features and deep learning according to claim 1, characterized in that: In step four, the preprocessing operations of the sound signal include noise reduction and feature extraction; the noise reduction is preferably performed using wavelet decomposition; the features are preferably short-time Fourier transform image features.

6. The detection method of the timber structure health status detection device based on acoustic features and deep learning according to claim 1, characterized in that: In step four, the artificial intelligence algorithm is one of MobileNetV3, ShuffleNetV2, EfficientNet, GhostNet, and SqueezeNet lightweight convolutional neural networks to ensure efficient operation under limited computing resources. The artificial intelligence algorithm is improved by introducing transfer learning technology and attention mechanism, thereby enhancing the algorithm's transfer learning ability and detection accuracy.

7. The detection method of the timber structure health status detection device based on acoustic features and deep learning according to claim 6, characterized in that: The transfer learning technique involves first pre-training the algorithm on the large-scale general dataset Wikipedia, and then using the pre-trained algorithm to train and fine-tune the algorithm's parameters on the timber structure detection data; the attention mechanism is the Squeeze-and-Excitation (SE) module.

8. The detection method of the timber structure health status detection device based on acoustic features and deep learning according to claim 1, characterized in that: In step four, the artificial intelligence algorithm needs to be trained indoors in advance. The training dataset should take into account different variables such as defect type, defect size, defect location, striking force, striking location, structural form, structural size, and structural stress to ensure the generalization of the artificial intelligence algorithm.

9. The detection method of the timber structure health status detection device based on acoustic features and deep learning according to claim 1, characterized in that: In step four, the health status test result index of the wooden structure under test is the damage probability.

10. The detection method of the timber structure health status detection device based on the combination of acoustic features and deep learning according to claim 1, characterized in that: In step six, the tools for drawing contour maps are Oringin, Matalab, or Python software. The color scheme of the contour maps should gradually change from blue to red as different damage probability ranges increase. The higher the damage probability value, the redder the color, indicating that the area is more likely to be damaged, thereby obtaining the health status of different areas of the wooden structure.

Images