Portable bolt looseness detection device and detection method
Through the portable bolt loosening detection device, the acoustic wave signal is analyzed using signal processing and LSTM model, and the problem of environmental noise and sample collection difficulties in soundprint signal detection is solved, efficient and accurate bolt loosening detection is achieved, and detection accuracy and efficiency are improved.
Patent Information
- Application Number
- CN202411278967.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-12
- Publication Date
- 2024-12-06
AI Technical Summary
In the prior art, voiceprint signal detection has problems such as large environmental noise influence, difficulty in collecting fault samples, and difficulty in extracting voiceprint signal characteristics of equipment in bolt loosening detection, resulting in low detection accuracy.
Design a portable bolt loose detection device, including a human-computer interactive host, acoustic wave transmitter and multiple acoustic wave sensors, use a signal processing module to perform noise reduction processing, and analyze the sound wave signal through the LSTM neural network model to determine whether the bolt is loose. The device is built-in Filter circuit, amplifier circuit and analog-to-digital conversion circuit, the frequency of the acoustic wave transmitter is adjustable, and signal processing is performed with STFT, spectrogram analysis and DSP soundprint noise reduction algorithm.
It significantly improves detection accuracy and efficiency, can accurately identify bolt looseness, reduce equipment failures and safety accidents, and the automated inspection process greatly improves detection speed and accuracy, and improves detection accuracy by more than 30%.
Smart Images

Figure CN119086714A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of transmission tower bolt loosening fault diagnosis, and in particular to a portable bolt loosening detection device and a detection method. Background Art
[0002] As an indispensable part of mechanical equipment maintenance, bolt loosening detection is of great importance. In a complex and ever-changing industrial environment, mechanical equipment is often subjected to huge workloads and vibration shocks. These external factors can easily lead to loosening of bolted joints. Once the loose bolts are not discovered and handled in time, they may cause the dislocation of equipment components, increased wear, and even the collapse of the entire mechanical system, leading to serious equipment failures and safety accidents, causing huge losses to production activities, and also threatening the lives of operators.
[0003] Therefore, exploring efficient and accurate bolt loosening detection methods is extremely important to ensure the safe operation of mechanical equipment. As an important technology in the field of modern non-destructive testing, acoustic wave detection technology has been widely used in bolt loosening detection in recent years and has shown significant advantages.
[0004] Acoustic wave detection technology uses the principle that when sound waves propagate in the medium, they will produce physical phenomena such as reflection, refraction, and scattering when encountering different interfaces or defects. By measuring and analyzing the changes in these sound wave signals, non-contact and rapid detection of the bolt connection status can be achieved. In bolt loosening detection, acoustic wave detection technology can capture the tiny vibration or acoustic emission signals generated when the bolts are loose, and these signals contain key information about the loosening status of the bolts. By accurately analyzing and processing these signals, it is possible to accurately determine whether the bolts are loose, as well as the degree and location of the looseness. Compared with traditional detection methods such as visual inspection, tapping and listening, acoustic wave detection technology has higher detection accuracy and reliability, can promptly detect potential bolt loosening problems, and effectively prevent equipment failures and safety accidents caused by bolt loosening. At the same time, acoustic wave detection technology also has the advantages of non-contact, non-destructive, and rapid, and can perform detection without affecting the normal operation of mechanical equipment, improving detection efficiency and convenience.
[0005] The application of acoustic wave detection technology in bolt loosening detection can refer to the following patent documents:
[0006] Reference 1: Chinese patent document with application number 202310811971.3
[0007] Reference 1 records a method and device for diagnosing loose screws on a tower based on voiceprint recognition, which relates to the field of fault diagnosis technology. The real-time tower information of the target tower is obtained, and the target sound wave signal generated by the sound wave generator is adaptively determined according to the real-time tower information; the target sound wave signal is emitted to the target tower, and the soundprint signals at each screw connection of the target tower are collected; for the soundprint signal at each screw connection, the MFCC feature vector of the soundprint signal is extracted, and the MFCC feature vector is weighted and optimized to obtain the target feature vector; the target feature vector is input into a preset acoustic model to obtain the diagnosis and identification result of whether the screw is loose. An adaptive active sound system is adopted, and the sound content is automatically analyzed and adapted according to the tower model, size, environment and other conditions, which greatly optimizes the efficiency of soundprint analysis, and the preset acoustic model is used to efficiently scan and analyze the tightening status of the bolts of the entire tower, determine the looseness of the bolts, and improve the detection accuracy and work efficiency.
[0008] Reference 2: Chinese patent document with application number 202410616950.0
[0009] Reference 2 records a method for detecting loose tower bolts based on voiceprint signals, including the following steps: S1 collects audio samples of tower bolts and processes the data to generate a tower bolt dataset; S2 combines the processed tower bolt dataset to build a graph convolutional neural network and multi-domain adaptive model to obtain bolt feature embedding; S3 randomly extracts data from the tower bolt dataset to train and update the graph convolutional network and multi-domain adaptive model; S4 introduces prediction target constraints to the trained graph convolutional network and multi-domain adaptive model to output the overall tower bolt working condition. A multi-channel domain graph convolutional network method is used to extract the common features of bolts in the source domain and the target domain and to mine the relationship between the features, thereby reducing the distribution difference between the target domain bolt data and the source domain bolt data, and solving the problem of low accuracy caused by insufficient data in the prior art.
[0010] However, bolt loosening detection based on voiceprint signals still has the following problems: (1) Since various environmental noises have a great impact on the equipment status information carried in the voiceprint signal, how to reduce noise is the primary problem that needs to be solved; (2) It is difficult to collect positive samples of industrial equipment failures: It takes months or even years to collect fault noise, and it is difficult to form a large fault noise training set; (3) The algorithm for extracting the characteristics of equipment voiceprint signals is difficult: Using parameter recognition algorithms, it is necessary to independently establish parameter values corresponding to faults of various types of equipment for various grades and models of equipment, and the parameter table must be reconstructed every time a device is replaced. Summary of the invention
[0011] The purpose of the present invention is to solve the above-mentioned technical problems existing in the prior art and to provide a portable bolt loosening detection device and detection method.
[0012] In order to solve the above technical problems, the present invention adopts the following technical solutions: a portable bolt loosening detection device, comprising a box body, a human-machine interactive host built in the box body, a sound wave transmitter and a plurality of sound wave sensors;
[0013] The human-computer interaction host includes a display screen and operation buttons, and the human-computer interaction host has a built-in signal processing module and a data analysis module;
[0014] The sound wave transmitter can be installed on the iron tower to be detected, and is used to generate and transmit sound waves of a specific frequency;
[0015] The acoustic wave sensor can be installed at the bolt to be tested and connected to the human-machine interaction host through a cable to capture the reflected acoustic wave generated by the bolt;
[0016] The signal processing module is used to receive the sound wave signal transmitted by the sound wave sensor and perform noise reduction processing on the signal to obtain a preprocessed sound wave signal;
[0017] The data analysis module is used to receive the pre-processed sound wave signal transmitted by the signal processing module, analyze the signal through the trained neural network model, and determine whether the bolt is loose.
[0018] As a further optimization of the portable bolt loosening detection device of the present invention: the emission frequency of the sound wave transmitter is adjustable to meet the detection requirements of bolts of different materials and sizes.
[0019] As a further optimization of the portable bolt loosening detection device of the present invention: the signal processing module includes a filtering circuit, an amplifying circuit and an analog-to-digital conversion circuit.
[0020] As a further optimization of the portable bolt loosening detection device of the present invention: the neural network model adopted by the data analysis module is an LSTM network model.
[0021] The present invention also provides a method for detecting bolt loosening, which uses the above detection device to perform detection operations, specifically comprising the following steps:
[0022] S1. Install the acoustic wave sensor at the predetermined monitoring bolt position, ensuring that the contact surface between the sensor and the bolt is flat to obtain accurate acoustic wave data;
[0023] S2. Connect the human-machine interaction host and the acoustic wave sensor through a dedicated cable to realize signal transmission between the acoustic wave sensor and the signal processing module, ensure a stable connection, and avoid interference during signal transmission;
[0024] S3. Start the human-machine interaction host and preheat it to ensure that all components are in a stable working state. Set the detection parameters through the built-in software of the human-machine interaction host and start the detection. The sound wave transmitter will emit sound waves of a specific frequency. The sound wave sensor receives the sound wave signal generated by the loose bolt and transmits the signal to the signal processing module.
[0025] S4. The signal processing module performs noise reduction processing on the received sound wave signal and transmits it to the data analysis module. The data analysis module analyzes the processed signal, identifies abnormal signal data, determines whether the bolt is loose, generates a test report and displays it on the screen of the human-computer interaction host.
[0026] As a further optimization of the bolt loosening detection method of the present invention: the detection parameter in step S3 includes the sound wave output frequency of the sound wave transmitter, and the sound wave output frequency is determined by the following formula:
[0027] f=f0×(1+α×ΔT+β×V w +γ×M)
[0028] Where f is the sound wave output frequency, f0 is the basic frequency, ΔT is the temperature change value, V w is the ambient wind speed, M is the matrix material factor, and α, β and γ are factor coefficients.
[0029] As a further optimization of the bolt loosening detection method of the present invention: a mathematical model is established to describe the relationship between the sound wave frequency and environmental factors, the model includes factor coefficients α, β, γ and M, the experiment is repeated and the stability of the coefficients is verified, and the model is fitted through experimental data to determine the specific values of the coefficients α, β, γ and M.
[0030] As a further optimization of the bolt loosening detection method of the present invention: the method of performing noise reduction processing on the signal in step S4 is:
[0031] Use STFT to divide the sound wave signal into multiple short time windows, perform Fourier transform on each time window, obtain the spectrum information of the signal in the window, combine the spectrum information of all time windows to form the time-frequency representation of the sound wave signal, and extract the sound wave unit with unique voiceprint characteristics;
[0032] Using the spectrogram analysis algorithm, the spectrum of the sound wave signal is divided into different frequency bands or frequency bands. In each frequency band or frequency band, the sound wave units with specific voiceprint characteristics are identified to achieve the stratification of the sound wave;
[0033] The DSP voiceprint denoising algorithm, the spectrogram analysis algorithm and STFT are combined to form a comprehensive sound wave processing system. The system first uses STFT to segment and analyze the sound wave signal, then uses the spectrogram analysis algorithm to segment the spectrum and identify the voiceprint units, and finally uses the DSP voiceprint denoising algorithm to suppress the identified noise.
[0034] As a further optimization of the bolt loosening detection method of the present invention: the method in step S4 for the data analysis module to analyze the processed signal and identify abnormal signal data is:
[0035] Collect the acoustic wave signal data of the transmission tower bolts in different working conditions, the acoustic wave signal data includes the specific sound pattern feature data generated when the bolts are loose;
[0036] The first voiceprint information is filtered using a Mel filter to obtain a Mel domain spectrum, the Mel domain spectrum is logarithmic, and then a discrete cosine transform is performed to obtain an MFCC feature vector, and the MFCC feature vector is used as the input of the LSTM model to construct an LSTM network, and the advanced characteristics of the attention mechanism are combined to improve the modeling ability of the sound wave sequence;
[0037] The LSTM model is trained using preprocessed sound wave data to enable it to identify and classify different voiceprint patterns. In the early stages of training, manually labeled positive samples are used for 1:1 screening to ensure the accuracy of model learning. As model training progresses, the model is used to predict unlabeled data, and high-confidence prediction results are added to the training set as new training samples to achieve 1:N automatic labeling. At the same time, a user feedback system is established to allow users to independently label and submit to the system when they find fault types that the model fails to identify;
[0038] The acoustic wave signals collected and processed in real time during the detection process are input into the trained LSTM model, and the output results are used to determine whether the bolts are loose.
[0039] As a further optimization of the bolt loosening detection method of the present invention: the structure of the LSTM model is as follows:
[0040] Input layer: receives preprocessed sound wave data;
[0041] LSTM layer: You can set multiple LSTM layers, each containing the following parameters: input_size: the dimension of the feature; hidden_size: the size of the hidden layer; num_layers: the number of LSTM layers; batch_first: set to True; dropout: If necessary, dropout can be added between LSTM layers to prevent overfitting;
[0042] Fully connected layer: After the LSTM layer, one or more fully connected layers can be added to map the output of the LSTM to the final prediction target;
[0043] Output layer: Depending on the task, the output layer can be one or more neurons. The number of neurons in the output layer is equal to the number of categories, and the softmax activation function is usually used for multi-classification;
[0044] The input data is processed by the LSTM layer. Each LSTM unit processes the sequence data and generates an output at each time step. The output of the LSTM layer can be the hidden state of the last time step or the set of hidden states of all time steps. The output of the LSTM layer is converted into the final result through the fully connected layer.
[0045] The present invention has the following beneficial effects:
[0046] 1. The present invention uses advanced acoustic wave detection technology to accurately identify small changes in bolt loosening, thereby significantly improving detection accuracy. The device uses the propagation characteristics of sound waves in bolts to accurately determine the tightening state of the bolts by analyzing the reflection and propagation time of the sound waves. At the same time, a machine learning algorithm is used to analyze the collected acoustic wave data, and the characteristics of bolt loosening are identified by training the model, thereby improving the accuracy and reliability of detection. A large number of tests have been carried out in laboratory and field environments. Compared with traditional manual detection methods, the detection accuracy of the device has been improved by more than 30%, effectively reducing equipment failures and safety accidents caused by loose bolts.
[0047] 2. The present invention realizes an automated and intelligent detection process, greatly improving the detection efficiency. The device can automatically collect and analyze acoustic wave data without human intervention, greatly improving the detection speed and efficiency. The device adopts a portable design, which is easy to carry and operate, allowing inspectors to quickly deploy and use it in various environments, further saving detection time. The device has a built-in intelligent diagnostic system that can provide real-time detection results and maintenance recommendations, helping users to take appropriate maintenance measures in a timely manner, avoiding downtime and maintenance costs caused by loose bolts. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 It is a structural block diagram of the bolt loosening detection device of the present invention. DETAILED DESCRIPTION
[0049] In order to better understand the present invention, the content of the present invention is further explained below in conjunction with embodiments, but the content of the present invention is not limited to the following embodiments.
[0050] <Bolt loosening detection device>
[0051] A portable bolt loosening detection device includes a box body and a human-machine interaction host, a sound wave transmitter and a plurality of sound wave sensors built in the box body. The human-machine interaction host includes a display screen and operation buttons, and the human-machine interaction host is built with a signal processing module and a data analysis module. The sound wave transmitter can be installed on the iron tower to be detected, and is used to generate and transmit sound waves of a specific frequency. The sound wave sensor can be installed at the bolt to be tested, and is connected to the human-machine interaction host through a cable, and is used to capture the reflected sound waves generated by the bolt. The signal processing module is used to receive the sound wave signal transmitted by the sound wave sensor, and perform noise reduction processing on the signal to obtain a preprocessed sound wave signal. The data analysis module is used to receive the preprocessed sound wave signal transmitted by the signal processing module, analyze the signal through a trained neural network model, and determine whether the bolt is loose.
[0052] The frequency of the sound wave emitted by the sound wave transmitter is adjustable to meet the needs of bolt detection of different materials and sizes. The signal processing module includes a filter circuit, an amplifier circuit, and an analog-to-digital conversion circuit. The neural network model used in the data analysis module is the LSTM network model.
[0053] <Bolt loosening detection method>
[0054] A method for detecting bolt loosening comprises the following steps:
[0055] S1. Install the acoustic wave sensor at the predetermined monitoring bolt position and ensure that the contact surface between the sensor and the bolt is flat to obtain accurate acoustic wave data.
[0056] S2. Connect the human-computer interaction host and the acoustic wave sensor through a dedicated cable to realize signal transmission between the acoustic wave sensor and the signal processing module, ensure a stable connection, and avoid interference during signal transmission.
[0057] S3. Start the human-machine interaction host and preheat it to ensure that all components are in a stable working state. Set the detection parameters through the built-in software of the human-machine interaction host and start the detection. The sound wave transmitter will emit sound waves of a specific frequency. The sound wave sensor receives the sound wave signal generated by the loose bolts and transmits the signal to the signal processing module.
[0058] The detection parameters include the sound wave output frequency of the sound wave transmitter, which is determined by the following formula:
[0059] f=f0×(1+α×ΔT+β×Vw+γ×M)
[0060] Among them, f is the sound wave output frequency, f0 is the basic frequency, ΔT is the temperature change value, Vw is the ambient wind speed, M is the matrix material factor, and α, β and γ are all factor coefficients.
[0061] A mathematical model is established to describe the relationship between sound wave frequency and environmental factors. The model contains factor coefficients α, β, γ and M. The experiment is repeated and the stability of the coefficients is verified. The model is fitted with experimental data to determine the specific values of the coefficients α, β, γ and M.
[0062] S4. The signal processing module performs noise reduction processing on the received sound wave signal and transmits it to the data analysis module. The data analysis module analyzes the processed signal, identifies abnormal signal data, determines whether the bolt is loose, generates a test report and displays it on the screen of the human-computer interaction host.
[0063] The method of noise reduction for the signal is:
[0064] Use STFT to divide the sound wave signal into multiple short time windows, perform Fourier transform on each time window, obtain the spectrum information of the signal in the window, combine the spectrum information of all time windows to form the time-frequency representation of the sound wave signal, and extract the sound wave unit with unique voiceprint characteristics;
[0065] Using the spectrogram analysis algorithm, the spectrum of the sound wave signal is divided into different frequency bands or frequency bands. In each frequency band or frequency band, the sound wave units with specific voiceprint characteristics are identified to achieve the stratification of the sound wave;
[0066] The DSP voiceprint denoising algorithm, the spectrogram analysis algorithm and STFT are combined to form a comprehensive sound wave processing system. The system first uses STFT to segment and analyze the sound wave signal, then uses the spectrogram analysis algorithm to segment the spectrum and identify the voiceprint units, and finally uses the DSP voiceprint denoising algorithm to suppress the identified noise.
[0067] The method by which the data analysis module analyzes the processed signal and identifies abnormal signal data is as follows:
[0068] Collect the acoustic wave signal data of the transmission tower bolts in different working conditions, the acoustic wave signal data includes the specific sound pattern feature data generated when the bolts are loose;
[0069] The first voiceprint information is filtered using a Mel filter to obtain a Mel domain spectrum, the Mel domain spectrum is logarithmic, and then a discrete cosine transform is performed to obtain an MFCC feature vector, and the MFCC feature vector is used as the input of the LSTM model to construct an LSTM network, and the advanced characteristics of the attention mechanism are combined to improve the modeling ability of the sound wave sequence;
[0070] The LSTM model is trained using preprocessed sound wave data to enable it to identify and classify different voiceprint patterns. In the early stages of training, manually labeled positive samples are used for 1:1 screening to ensure the accuracy of model learning. As model training progresses, the model is used to predict unlabeled data, and high-confidence prediction results are added to the training set as new training samples to achieve 1:N automatic labeling. At the same time, a user feedback system is established to allow users to independently label and submit to the system when they find fault types that the model fails to identify;
[0071] The acoustic wave signals collected and processed in real time during the detection process are input into the trained LSTM model, and the output results are used to determine whether the bolts are loose.
[0072] The structure of the LSTM model is as follows:
[0073] Input layer: receives preprocessed sound wave data;
[0074] LSTM layer: You can set multiple LSTM layers, each containing the following parameters: input_size: the dimension of the feature; hidden_size: the size of the hidden layer; num_layers: the number of LSTM layers; batch_first: set to True; dropout: If necessary, dropout can be added between LSTM layers to prevent overfitting;
[0075] Fully connected layer: After the LSTM layer, one or more fully connected layers can be added to map the output of the LSTM to the final prediction target;
[0076] Output layer: Depending on the task, the output layer can be one or more neurons. The number of neurons in the output layer is equal to the number of categories, and the softmax activation function is usually used for multi-classification;
[0077] The input data is processed by the LSTM layer. Each LSTM unit processes the sequence data and generates an output at each time step. The output of the LSTM layer can be the hidden state of the last time step or the set of hidden states of all time steps. The output of the LSTM layer is converted into the final result through the fully connected layer.
[0078] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art may make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.
Claims
1. A portable bolt loosening detection device, Features: It includes a box body, a human-machine interaction host built in the box body, a sound wave transmitter and a plurality of sound wave sensors; The human-computer interaction host includes a display screen and operation buttons, and the human-computer interaction host has a built-in signal processing module and a data analysis module; The sound wave transmitter can be installed on the iron tower to be detected, and is used to generate and transmit sound waves of a specific frequency; The acoustic wave sensor can be installed at the bolt to be tested and connected to the human-machine interaction host through a cable to capture the reflected acoustic wave generated by the bolt; The signal processing module is used to receive the sound wave signal transmitted by the sound wave sensor and perform noise reduction processing on the signal to obtain a preprocessed sound wave signal; The data analysis module is used to receive the pre-processed sound wave signal transmitted by the signal processing module, analyze the signal through the trained neural network model, and determine whether the bolt is loose.
2. The portable bolt loosening detection device as claimed in claim 1, Features: The frequency of the sound waves emitted by the sound wave transmitter is adjustable to meet the needs of detecting bolts of different materials and sizes.
3. The portable bolt loosening detection device as claimed in claim 1, Features: The signal processing module includes a filtering circuit, an amplifying circuit and an analog-to-digital conversion circuit.
4. The portable bolt loosening detection device as claimed in claim 1, Features: The neural network model used by the data analysis module is the LSTM network model.
5. A method for detecting loose bolts, It is characterized in that The detection operation is performed using the detection device described in any one of claims 1 to 4, specifically comprising the following steps: S1. Install the acoustic wave sensor at the predetermined monitoring bolt position, ensuring that the contact surface between the sensor and the bolt is flat to obtain accurate acoustic wave data; S2. Connect the human-machine interaction host and the acoustic wave sensor through a dedicated cable to realize signal transmission between the acoustic wave sensor and the signal processing module, ensure a stable connection, and avoid interference during signal transmission; S3. Start the human-machine interaction host and preheat it to ensure that all components are in a stable working state. Set the detection parameters through the built-in software of the human-machine interaction host and start the detection. The sound wave transmitter will emit sound waves of a specific frequency. The sound wave sensor receives the sound wave signal generated by the loose bolt and transmits the signal to the signal processing module. S4. The signal processing module performs noise reduction processing on the received sound wave signal and transmits it to the data analysis module. The data analysis module analyzes the processed signal, identifies abnormal signal data, determines whether the bolt is loose, generates a test report and displays it on the screen of the human-computer interaction host.
6. A method for detecting bolt loosening as claimed in claim 5, Features: The detection parameter in step S3 includes the sound wave output frequency of the sound wave transmitter, and the sound wave output frequency is determined by the following formula: f=f 0 ×(1+α×ΔT+β×V w +γ×M) Where, f is the sound wave output frequency, f 0 is the basic frequency, ΔT is the temperature change value, V w is the ambient wind speed, M is the matrix material factor, and α, β and γ are factor coefficients.
7. A method for detecting bolt loosening as claimed in claim 6, Features: A mathematical model is established to describe the relationship between sound wave frequency and environmental factors. The model contains factor coefficients α, β, γ and M. The experiment is repeated and the stability of the coefficients is verified. The model is fitted with experimental data to determine the specific values of the coefficients α, β, γ and M.
8. A method for detecting bolt loosening as claimed in claim 5, Features: The method for performing noise reduction processing on the signal in step S4 is: Use STFT to divide the sound wave signal into multiple short time windows, perform Fourier transform on each time window, obtain the spectrum information of the signal in the window, combine the spectrum information of all time windows to form the time-frequency representation of the sound wave signal, and extract the sound wave unit with unique voiceprint characteristics; Using the spectrogram analysis algorithm, the spectrum of the sound wave signal is divided into different frequency bands or frequency bands. In each frequency band or frequency band, the sound wave units with specific voiceprint characteristics are identified to achieve the stratification of the sound wave; The DSP voiceprint denoising algorithm, the spectrogram analysis algorithm and STFT are combined to form a comprehensive sound wave processing system. The system first uses STFT to segment and analyze the sound wave signal, then uses the spectrogram analysis algorithm to segment the spectrum and identify the voiceprint units, and finally uses the DSP voiceprint denoising algorithm to suppress the identified noise.
9. A method for detecting bolt loosening as claimed in claim 5, Features: The method in step S4 for the data analysis module to analyze the processed signal and identify abnormal signal data is: Collect the acoustic wave signal data of the transmission tower bolts in different working conditions, the acoustic wave signal data includes the specific sound pattern feature data generated when the bolts are loose; The first voiceprint information is filtered using a Mel filter to obtain a Mel domain spectrum, the Mel domain spectrum is logarithmic, and then a discrete cosine transform is performed to obtain an MFCC feature vector, and the MFCC feature vector is used as the input of the LSTM model to construct an LSTM network, and the advanced characteristics of the attention mechanism are combined to improve the modeling ability of the sound wave sequence; The LSTM model is trained using preprocessed sound wave data to enable it to identify and classify different voiceprint patterns. In the early stages of training, manually labeled positive samples are used for 1:1 screening to ensure the accuracy of model learning. As model training progresses, the model is used to predict unlabeled data, and high-confidence prediction results are added to the training set as new training samples to achieve 1:N automatic labeling. At the same time, a user feedback system is established to allow users to independently label and submit to the system when they find fault types that the model fails to identify; The acoustic wave signals collected and processed in real time during the detection process are input into the trained LSTM model, and the output results are used to determine whether the bolts are loose.
10. A method for detecting bolt loosening as claimed in claim 8, Features: The structure of the LSTM model is as follows: Input layer: receives preprocessed sound wave data; LSTM layer: You can set multiple LSTM layers, each containing the following parameters: input_size: the dimension of the feature; hidden_size: the size of the hidden layer; num_layers: the number of LSTM layers; batch_first: set to True; dropout: If necessary, dropout can be added between LSTM layers to prevent overfitting; Fully connected layer: After the LSTM layer, one or more fully connected layers can be added to map the output of the LSTM to the final prediction target; Output layer: Depending on the task, the output layer can be one or more neurons. The number of neurons in the output layer is equal to the number of categories, and the softmax activation function is usually used for multi-classification; The input data is processed by the LSTM layer. Each LSTM unit processes the sequence data and generates an output at each time step. The output of the LSTM layer can be the hidden state of the last time step or the set of hidden states of all time steps. The output of the LSTM layer is converted into the final result through the fully connected layer.
Citation Information
Cited By
Dynamic cable fastening state monitoring method based on optical fiber sensing
CN119377658A
A method for monitoring the fastening state of a dynamic cable based on optical fiber sensing
CN119377658B
Security monitoring automation control system applied to power grid power transmission tower
CN121441963A
Knife box bolt anti-loose monitoring device
CN224532733U
