Smart bracelet for sound anomaly detection and sound anomaly detection method

Abstract

The invention relates to the technical field of sound detection, in particular to a sound anomaly detection smart bracelet and a detection method, and a sound anomaly detection method. The smart bracelet for sound anomaly detection is matched with an analysis host and acoustic detection software, and an acoustic detection algorithm based on spectral analysis and machine learning is used for detecting the sound anomaly. And automatic detection and alarm of abnormal sound are realized. According to the invention, on-line automatic detection of wiring harness plugging is realized, the outdated condition of manual detection of an existing production line is replaced, the detection efficiency and accuracy are higher than those of manual detection, the SOP automation level is improved, and information archiving and backtracking are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of sound detection, and in particular to a sound abnormality detection smart bracelet and a detection method. Background Art

[0002] During the plugging process of the wiring harness of lithium batteries or electronic products, problems such as poor contact, incomplete plugging, and missing plugging may occur. Currently, the main method of detection is to manually pull back and draw lines, which not only affects the production rhythm, the detection effect is unstable, and the production cost is increased, but also affects the automation and information tracing of the production line. Therefore, it is necessary to realize automated sound detection to replace manual detection. Summary of the invention

[0003] The present invention provides a sound abnormality detection smart bracelet and a detection method, which can realize automatic detection and alarm of sound abnormality.

[0004] In order to achieve the purpose of the present invention, the technical solution adopted is: a sound anomaly detection method, a smart bracelet for sound anomaly detection cooperates with an analysis host and acoustic detection software, and an acoustic detection algorithm based on spectrum analysis and machine learning to realize automatic detection and alarm of sound anomalies.

[0005] As an optimization scheme of the present invention, the acoustic detection software includes a sound collection module, an acoustic analysis algorithm module, a visualization display module, an alarm output module and a communication linkage module. The acoustic analysis algorithm module executes an acoustic detection algorithm based on spectrum analysis and machine learning. The acoustic detection algorithm based on spectrum analysis and machine learning includes:

[0006] S11, frequency domain transformation: convert the audio signal into frequency domain through Fourier transform or wavelet transform;

[0007] S12, signal enhancement: enhance the signal of the target frequency band and suppress the noise signal;

[0008] S13, after signal enhancement, inverse transform from frequency domain to time domain;

[0009] S14, peak-valley analysis: in the time domain, determine whether the sound signal meets the preset target sound characteristics by identifying the local peaks and valleys of the signal;

[0010] S15, signal classification: The target sound is taken as a positive sample, and the non-target sound is taken as a negative sample. Features are extracted, and classified through a machine learning algorithm to train a target sound model. During detection, features of the candidate target sound are extracted and sent to the target sound model for detection. If it is the target sound, it is retained, otherwise it is ignored.

[0011] S16, statistical determination: counting the number of detected target sound segments.

[0012] As an optimized solution of the present invention, in step S11, for the discrete signal x[n], the discrete Fourier transform can be expressed as:

[0013]

[0014] where N is the length of the signal, k is the index of the frequency component, X[k] is the complex value corresponding to the frequency k, and n is the sampling time;

[0015] The definition of the continuous wavelet transform is:

[0016]

[0017] where ψ(t) is the mother wavelet function, a is the scale parameter, b is the translation parameter, and x(t) represents the original continuous-time signal.

[0018] As an optimized solution of the present invention, in step S12, a band-pass filter is used to enhance the signal in the target frequency band, and the transfer function of the band-pass filter is:

[0019]

[0020] where f1 and f2 are the lower and upper limit frequencies of the band-pass filter, respectively.

[0021] As an optimized solution of the present invention, in step S14, the local peak p[i] satisfies the condition:

[0022] x[i]>x[i―1] and x[i]>x[i+1]

[0023] The local valley v[i] satisfies:

[0024] x[i]<x[i―1] and x[i]<x[i+1]

[0025] By analyzing the amplitude and duration of the local peaks and valleys, it is determined whether the pattern of the sound signal conforms to the preset target sound characteristics.

[0026] As an optimized solution of the present invention, in step S15, Mel cepstral coefficients are used to extract features, specifically including:

[0027] 1) Filter the frequency-domain signal through a set of Mel filters to obtain the filtered energy:

[0028]

[0029] where X[k] is the frequency-domain signal, m is the index of the Mel filter, f1 is the index of the frequency-domain signal corresponding to the lowest frequency covered by the Mel filter, and f2 is the index of the frequency-domain signal corresponding to the highest frequency covered by the Mel filter;

[0030] 2) Take the logarithm of the filtered energy and apply discrete cosine transform to get MFCC:

[0031]

[0032] Among them, c n is the nth MFCC coefficient, and M is the total number of Mel filters.

[0033] As an optimization solution of the present invention, in step S15, the extracted features are classified using a machine learning algorithm, such as a support vector machine SVM, and also include a variety of other applicable machine learning algorithms. The SVM decision function is:

[0034]

[0035] Among them, α i is the Lagrange multiplier, y i is the sample label, K(x i ,x) is the kernel function, b is the bias term; the classification result determines whether the sound signal belongs to the target sound or the abnormal sound.

[0036] In order to achieve the purpose of the present invention, the technical scheme adopted is: a smart bracelet for sound abnormality detection, including a wristband, a wireless sound sensor (including a wireless transmitter), a wireless sound receiver, a wireless vibration display (including a wireless receiver), a wireless vibration transmitter and a control device. The wireless sound sensor is fixed to the wristband by magnetic attraction and reinforced with a protective device to avoid collision and slipping. The wireless vibration display is fixed on the wristband. The wireless sound sensor is triggered by the control device to collect sound data, and the sound data is sent to the analysis host through the wireless transmitter. The analysis host automatically detects sound abnormalities and sends the detection results to the wireless receiver. The wireless vibration display vibrates and displays the number of detected targets when the detection result is a target sound.

[0037] As an optimization solution of the present invention, the analysis host drives the sound and light alarm when the detection result is the target sound.

[0038] As an optimization solution of the present invention, the control button is used to set the vibration cycle and display time of the wireless vibration display by short pressing.

[0039] The present invention has positive effects: 1) The present invention adopts wireless communication, convenient deployment, clear sound reception, automatic analysis, archiving, alarm, and communication to realize online automatic detection of wiring harness plug-in, replacing the backward situation of manual detection of existing production lines. The detection efficiency and accuracy are higher than manual detection, which improves the automation level of SOP and realizes information archiving and backtracking;

[0040] 2) The present invention replaces the manual back-pulling and drawing method with automated sound detection, significantly improving the detection efficiency on the production line. The automated system can quickly and accurately analyze a large amount of sound data and provide feedback in real time, reducing the possibility of human error. The acoustic detection algorithm based on spectrum analysis and machine learning can accurately identify normal plug-in sounds and abnormal sounds (such as poor contact, inadequate plugging, etc.), thereby improving the accuracy of the detection results;

[0041] 3) The design of the smart bracelet of the present invention allows it to be easily integrated into existing production lines without the need for large-scale transformation of existing equipment. This helps to achieve the digital transformation of production lines and improves the overall degree of automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0043] Figure 1 It is a schematic diagram of an embodiment of a smart bracelet for abnormal sound detection of the present invention having a wireless sound sensor;

[0044] Figure 2 It is a schematic diagram of an embodiment of a smart bracelet for abnormal sound detection of the present invention having a wireless vibration display;

[0045] Figure 3 It is a schematic diagram of another embodiment of the smart bracelet for abnormal sound detection of the present invention having a wireless sound sensor;

[0046] Figure 4 It is a schematic diagram of another embodiment of the smart bracelet for abnormal sound detection of the present invention having a wireless vibration display;

[0047] Figure 5 Schematic diagram of an embodiment of the detection method of the present invention;

[0048] Among them: 1. Wristband, 2. Wireless sound sensor, 3. Velcro, 4. Wireless vibration display, 5. Magnetic sheet. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] like Figure 1As shown, the present invention discloses a method for detecting sound anomalies. A smart bracelet for detecting sound anomalies cooperates with an analysis host and acoustic detection software, and an acoustic detection algorithm based on spectrum analysis and machine learning is used to realize automatic detection and alarm of sound anomalies.

[0051] Sound anomaly detection includes the following steps:

[0052] S1. The operator wears a smart bracelet for abnormal sound detection on his wrist;

[0053] S2. After the product is in place, start the test before the operator starts working;

[0054] S3, collect and analyze sound, and provide real-time vibration digital display prompts;

[0055] S4. After the operator completes the assembly work, the inspection stops;

[0056] S5, the analysis host outputs the analysis results and alarm information;

[0057] S6. Repeat S2-S5.

[0058] The acoustic detection software includes a sound collection module, an acoustic analysis algorithm module, a visualization module, an alarm output module, and a communication linkage module. The sound collection module converts the data received wirelessly into audio signals for processing by the acoustic analysis algorithm module. In order to allow users to intuitively understand and analyze the processing results, the visualization module presents the data obtained from the acoustic analysis in the form of charts and other forms. This not only helps non-professionals to quickly grasp the key points of information, but also provides a basis for experts to further study. Common visualization forms include waveform graphs, spectrum graphs, time domain / frequency domain statistical characteristic curves, etc. When the acoustic analysis algorithm detects an abnormal situation under preset conditions, the alarm output module will immediately trigger the corresponding alarm mechanism. The acoustic analysis algorithm module executes an acoustic detection algorithm based on spectrum analysis and machine learning. The acoustic detection algorithms based on spectrum analysis and machine learning include:

[0059] S11, frequency domain transformation: convert the audio signal into frequency domain through Fourier transform or wavelet transform;

[0060] In step S11, for the discrete signal x[n], the discrete Fourier transform can be expressed as:

[0061]

[0062] Where N is the length of the signal, k is the index of the frequency component, X[k] is the complex value corresponding to frequency k, and n is the sampling time. The Fourier transform can transform the time domain signal into a spectrum representation in the frequency domain, revealing the frequency components of the sound signal. By analyzing the amplitude and phase of different frequency components, the characteristics of the sound signal can be determined.

[0063] The definition of the continuous wavelet transform is as follows:

[0064]

[0065] Among them, ψ(t) is the mother wavelet function, a is the scale parameter, b is the translation parameter, and x(t) represents the original continuous-time signal. The wavelet transform decomposes the signal into wavelet functions of different scales to analyze the frequency and time information of the signal simultaneously. The wavelet transform is suitable for analyzing transient signals and can better capture the sudden changes in sound signals. The Fourier transform or the wavelet transform can be used simultaneously to transform the audio signal into the frequency domain.

[0066] S12. Signal enhancement: Enhance the signal in the target frequency band and suppress the noise signal, that is, suppress the irrelevant noise;

[0067] In step S12, a band-pass filter is used to enhance the signal in the target frequency band. That is, the main purpose is to weaken the low-frequency and high-frequency noises and enhance the intermediate-frequency signal of interest. The transfer function of the band-pass filter is as follows:

[0068]

[0069] Among them, f1 and g2 are the lower and upper limit frequencies of the band-pass filter respectively. The signal processed by the filter is more focused on the target frequency band, which helps to improve the accuracy of subsequent analysis.

[0070] S13. After signal enhancement, inverse transform from the frequency domain to the time domain;

[0071] S14. Peak-valley analysis: In the time domain, determine whether the sound signal conforms to the preset target sound characteristics by identifying the local peaks and valleys of the signal, that is, peak-valley detection. The core idea of peak-valley detection is to analyze the fluctuation characteristics by finding the local maximum (peak) and minimum (valley) values in the signal;

[0072] In step S14, the local peak p[i] satisfies the condition:

[0073] x[i]>x[i - 1] and x[i]>x[i + 1]

[0074] The local valley v[i] satisfies:

[0075] x[i]<x[i - 1] and x[i]<x[i + 1]

[0076] Judge whether the pattern of the sound signal conforms to the preset target sound characteristics by analyzing the amplitude and duration of the local peaks and valleys.

[0077] S15, signal classification: The target sound is taken as a positive sample, and the non-target sound is taken as a negative sample. Features are extracted, and classified through a machine learning algorithm to train a target sound model. During detection, features of the candidate target sound are extracted and sent to the target sound model for detection. If it is the target sound, it is retained, otherwise it is ignored.

[0078] In step S15, Mel-frequency cepstral coefficients are used to extract features. The feature extraction method Mel-frequency cepstral coefficients (MFCC) simulates the human ear's perception of different frequencies and performs nonlinear transformation on the spectrum. Specifically, it includes:

[0079] 1) Filter the frequency domain signal through a set of Mel filters to obtain the filtered energy:

[0080]

[0081] Where X[k] is the frequency domain signal, m is the index of the Mel filter, f1 is the index of the frequency domain signal corresponding to the lowest frequency covered by the Mel filter, and f2 is the index of the frequency domain signal corresponding to the highest frequency covered by the Mel filter;

[0082] 2) Take the logarithm of the filtered energy and apply discrete cosine transform to get MFCC:

[0083]

[0084] Among them, c n is the nth MFCC coefficient, and M is the total number of Mel filters.

[0085] For the extracted features, a machine learning algorithm is used to classify the sound signal, such as a support vector machine (SVM), and also includes a variety of other applicable machine learning algorithms. The SVM decision function is:

[0086]

[0087] Among them, α i is the Lagrange multiplier, y i is the sample label, K(x i ,x) is the kernel function, b is the bias term; the classification result determines whether the sound signal belongs to the target sound (such as the normal sound of the wiring harness plug-in) or the abnormal sound (such as the sound of poor contact).

[0088] S16, statistical determination: Count the number of detected target sound segments. If it is consistent with the set parameters, it is judged as OK, otherwise it is judged as NG.

[0089] A smart bracelet for detecting sound anomalies, including a wristband, a wireless sound sensor (including a wireless transmitter), a wireless sound receiver, a wireless vibration display (including a wireless receiver), a wireless vibration transmitter and a control device, wherein the wireless sound sensor is fixed on the wristband by magnetic attraction, and the wireless vibration display is fixed on the wristband. The wireless sound sensor is triggered by the control device to collect sound data, and the sound data is sent to the analysis host through the wireless transmitter. The analysis host automatically detects sound anomalies, and sends the detection results to the wireless vibration display through the wireless vibration transmitter. When the detection result is a target sound, the display vibrates and displays the number of detected targets. Among them, the wristband is made of silicone, woven fabric or metal, and is fixed by snap or magnetic attraction (magnetic attraction sheet 5). The wireless sound sensor includes a microphone, which can be omnidirectional or cardioid. The wireless sound receiver and the wireless transmitter use LDS high-gain antenna, GFSK wireless modulation mode, and 2.4G adaptive frequency hopping. The control button of the sensor includes a switch and a noise reduction function. The switch can be started or stopped by long pressing, and the noise reduction can be started or stopped by short pressing. The noise reduction function filters high-frequency noise. The wireless vibration display includes a vibrator, a display and a button. When the detection result is a target sound, the vibrator vibrates and displays the number of detected targets through the display. The smart bracelet for abnormal sound detection also includes a charging device, which is a charging box and a charging cable for charging the wireless sound sensor and the wireless sound receiver, the wireless vibration display and the wireless vibration transmitter.

[0090] The analysis host drives the sound and light alarm when the detection result is the target sound. The analysis host is a hardware processing device such as an industrial computer, an embedded motherboard or a mobile phone. It has a standard audio input interface that can be connected to a wireless sound receiver. It also has a multi-function IO module that can drive the sound and light alarm, a wireless vibration transmitter and a communication module to communicate with the release control system on the production line and send the analysis result signal.

[0091] The control button can be pressed briefly to set the vibration cycle and display time of the wireless vibration display.

[0092] Embodiment 1,

[0093] The present invention provides a smart bracelet for detecting abnormal sound. Figure 1 , 2 , including the following components:

[0094] The wristband 1, wireless sound sensor 2, wireless sound receiver, wireless vibration display, wireless vibration transmitter, charging device, and control device are wireless buttons, and there is also a matching industrial computer as an analysis host to run the acoustic detection software. The wristband is made of Velcro 3, the wireless sound sensor 2 is fixed to the wristband by magnetic attraction, the wireless button is installed next to the operating table, and the wireless sound receiver is connected to the microphone interface of the industrial computer through an audio plug. The wireless vibration display 4 is fixed on the wristband, and the wireless vibration transmitter is connected to the multi-function IO module of the industrial computer.

[0095] The test consists of the following steps, see Figure 5 :

[0096] S1, the operator wears a smart bracelet on his wrist;

[0097] S2. Before the operator starts the operation, he starts the test by pressing the wireless button;

[0098] S3, receiving the start detection signal, collecting and analyzing the sound, and vibrating the digital display in real time to prompt;

[0099] S4. After the operator completes the assembly work, he stops the test by pressing the wireless button;

[0100] S5, the analysis host outputs the sound analysis results and alarm information;

[0101] S6. Repeat S2-S5.

[0102] In this embodiment, the sound analysis results are displayed on the software interface, and the audio file and visual photo file are saved at the same time. The sound and light alarm is controlled, and the green light turns on if the detection is OK, and the red light turns on if the detection is NG, and a sound alarm is issued. In addition, the results are sent to the production line host computer to control whether the product can be released to the next workstation.

[0103] Embodiment 2,

[0104] Different from the first embodiment, in this embodiment, the wristband is made of food-grade silicone material, which is soft and hypoallergenic and suitable for long-term wear. The wireless sound sensor 2 and the control device are both stuck in the corresponding hollow position of the wristband. Figure 3 ; The wireless vibration display 4 is stuck in the corresponding hollow position of the wristband, see Figure 4 ; No manual operation is required to start the inspection. When the product reaches the inspection station through AGV or conveyor belt, or the barcode reader reads the product code, the inspection software automatically reads the arrival signal and automatically starts the inspection. When the operation is completed, there is still manual operation to press the stop inspection button on the wireless button.

[0105] Embodiment 3,

[0106] Different from the first and second embodiments, in this embodiment, no control device is required, and the detection is started and stopped by voice control.

[0107] Embodiment 4,

[0108] Different from Embodiments 1, 2 and 3, in this embodiment, a wireless vibration display is not used, but a wireless vibrator is used instead, which reduces the size of the device and saves costs.

[0109] Embodiment 5,

[0110] Different from Embodiments 1, 2, and 3, in this embodiment, the wireless sound sensor, the control device, and the wireless vibration display are fixed on the same wristband, and the above communication method adopts the Bluetooth communication method, and other wireless communication methods such as WIFI and 5G may also be adopted.

[0111] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting abnormal sound, characterized in that: Through an intelligent bracelet for abnormal sound detection, an analysis host, and acoustic detection software, an acoustic detection algorithm based on spectrum analysis and machine learning is used to achieve automatic detection and alarm of abnormal sounds.

2. The method for detecting abnormal sound according to claim 1, characterized in that: The acoustic detection software includes a sound acquisition module, an acoustic analysis algorithm module, a visualization display module, an alarm output module, and a communication linkage module. The acoustic analysis algorithm module executes an acoustic detection algorithm based on spectrum analysis and machine learning. The acoustic detection algorithm based on spectrum analysis and machine learning includes: S11. Frequency-domain transformation: Transform the audio signal to the frequency domain through Fourier transform or wavelet transform. S12. Signal enhancement: Enhance the signal in the target frequency band and suppress the noise signal. S13. After signal enhancement, inverse transform from the frequency domain to the time domain. S14. Peak-valley analysis: In the time domain, determine whether the sound signal conforms to the preset target sound characteristics by identifying the local peaks and valleys of the signal. S15. Signal classification: Use the target sound as the positive sample and the non-target sound as the negative sample, extract features, and classify them through machine learning algorithms to train a target sound model. During detection, extract the features of the candidate target sound and send them to the target sound model for detection. If it belongs to the target sound, retain it; otherwise, ignore it. S16. Statistical determination: Statistically count the number of detected target sound segments.

3. The method for detecting abnormal sound according to claim 2, characterized in that: In step S11, for the discrete signal x[n], the discrete Fourier transform can be expressed as: where N is the length of the signal, k is the index of the frequency component, X[k] is the complex value corresponding to the frequency k, and n is the sampling time. The definition of the continuous wavelet transform is: where ψ(t) is the mother wavelet function, a is the scale parameter, b is the translation parameter, and x(t) represents the original continuous-time signal.

4. The method for detecting abnormal sound according to claim 3, characterized in that: In step S12, use a band-pass filter to enhance the signal in the target frequency band. The transfer function of the band-pass filter is: where f1 and f2 are the lower and upper limit frequencies of the band-pass filter, respectively.

5. The method for detecting abnormal sound according to claim 4, characterized in that: In step S14, the local peak p[i] satisfies the condition: x[i]>x[i―1] and x[i]>x[i+1] The local valley v[i] satisfies: x[i]<x[i―1] and x[i]<x[i+1] Judge whether the mode of the sound signal conforms to the preset target sound characteristics by analyzing the amplitude and duration of the local peaks and valleys.

6. The method for detecting abnormal sound according to claim 5, characterized in that: In step S15, use Mel-frequency cepstral coefficients to extract features, specifically including: 1) Filter the frequency-domain signal through a set of Mel filters to obtain the filtered energy: where X[k] is the frequency-domain signal, m is the index of the Mel filter, f1 is the index of the frequency-domain signal corresponding to the lowest frequency covered by the Mel filter, and f2 is the index of the frequency-domain signal corresponding to the highest frequency covered by the Mel filter. 2) Take the logarithm of the filtered energy and apply the discrete cosine transform to obtain the MFCC. Among them, c n is the nth MFCC coefficient, and M is the total number of Mel filters.

7. The method for detecting abnormal sound according to claim 5, characterized in that: In step S15, for the extracted features, use the machine learning algorithm support vector machine (SVM) to classify the sound signal. The SVM decision function is: Among them, α i is the Lagrange multiplier, y i is the sample label, K(z i ,x) is the kernel function, b is the bias term; the classification result determines whether the sound signal belongs to the target sound or the abnormal sound.

8. A smart bracelet for detecting abnormal sound, characterized in that: It includes a wristband, a wireless sound sensor, a wireless sound receiver, a wireless vibration display, a wireless vibration transmitter and a control device. The wireless sound sensor contains a wireless transmitter, and the wireless vibration display contains a wireless receiver. The wireless sound sensor is fixed to the wristband by magnetic attraction and reinforced with a protective device to prevent collision and slipping. The wireless vibration display is fixed to the wristband. The wireless sound sensor is triggered by the control device to collect sound data, and the sound data is sent to the analysis host through the wireless transmitter. The analysis host automatically detects sound abnormalities and sends the detection results to the wireless vibration display through the wireless vibration transmitter. When the detection result is the target sound, the wireless vibration display vibrates and displays the number of detected targets.

9. The smart bracelet for abnormal sound detection according to claim 8, characterized in that: The analysis host drives the sound and light alarm when the detection result is the target sound.

10. The smart bracelet for abnormal sound detection according to claim 9, characterized in that: The control button can be pressed briefly to set the vibration cycle and display time of the wireless vibration display.

Images