Abnormal sound detection method and abnormal sound detection device

Abstract

The present invention proposes an abnormal noise detection method and an abnormal noise detection device, which relate to the technical field of abnormal noise detection of fans, and are used to detect noise on the fan of a product. The abnormal noise detection method includes: collecting the noise signal generated when the fan is working; sequentially performing time-domain discontinuous abnormal noise signal recognition, abnormal signal recognition with relatively large background noise, and time-domain continuous abnormal noise signal recognition on the noise signal, and generating recognition results; and determining whether the fan has an abnormal noise problem based on the recognition results. This patent collects the fan noise signal, and sequentially performs time-domain discontinuous abnormal noise recognition, time-domain relatively large abnormal noise recognition, and time-domain continuous abnormal noise recognition on the noise signal generated by the fan, thereby determining whether the fan has an abnormal noise problem based on the recognition results, making product quality control more refined and improving the user experience.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormal noise detection of fans, and in particular to an abnormal noise detection method and an abnormal noise detection device. Background Art

[0002] Fan noise is an important quality indicator of products. At present, there are some methods on the market for identifying abnormal fan noise, such as interception using spectrum analysis, statistical characteristics analysis or manual listening. Due to the strong randomness of fan noise, poor spectrum consistency, and a wide variety of abnormal noises, the use of spectrum recognition methods has problems with high misjudgment and missed judgment rates. In addition, once the test equipment has an abnormality and does not have self-checking capabilities, it is easy to cause the failure of fan noise recognition for large quantities of products. The use of manual listening introduces subjectivity, resulting in inconsistent and unreliable recognition results. In addition, there are problems such as high labor costs, low efficiency, fatigue, and high error rates. Summary of the invention

[0003] The main purpose of the present invention is to provide an abnormal noise detection method and an abnormal noise detection device, aiming to more comprehensively and accurately identify abnormal fan noise of a product.

[0004] The present invention proposes an abnormal sound detection method for performing noise detection on a fan of a product, the method comprising: collecting a noise signal generated when the fan is working; sequentially performing time-domain discontinuous abnormal sound signal recognition, time-domain abnormal signal recognition with large background noise, and time-domain continuous abnormal sound signal recognition on the noise signal, and generating a recognition result; and determining whether the fan has an abnormal sound problem based on the recognition result.

[0005] In one embodiment, the noise signal is sequentially identified as a time-domain discontinuous abnormal sound signal, an abnormal signal with relatively large background noise, and an abnormal sound signal with relatively continuous time domain, and generating an identification result includes: obtaining the frequency band average energy variance of the noise signal, and identifying the noise signal as a time-domain discontinuous abnormal sound signal based on the frequency band average energy variance; obtaining a first average observation value of the noise signal, and identifying the noise signal as an abnormal signal with relatively large background noise based on the first average observation value; obtaining a second average observation value of the noise signal, and identifying the noise signal as a time-domain continuous abnormal sound signal based on the second average observation value.

[0006] In one embodiment, determining whether the fan has an abnormal noise problem based on the identification result includes: when the noise signal is identified as any one of an abnormal noise signal that is discontinuous in the time domain, an abnormal noise signal with a relatively large background noise, and an abnormal noise signal that is continuous in the time domain, determining that the fan has an abnormal noise problem; when the noise signal is identified as neither an abnormal noise signal that is discontinuous in the time domain, an abnormal noise signal with a relatively large background noise, nor an abnormal noise signal that is continuous in the time domain, determining that the fan does not have an abnormal noise problem.

[0007] In one embodiment, obtaining the frequency band average energy variance of the noise signal includes: collecting the noise signal of the fan, and obtaining a time domain signal based on the noise signal; performing high-pass filtering on the time domain signal to obtain a filtered time domain signal; obtaining time domain signals of multiple frames based on the filtered time domain signal; performing Fourier transform processing on the time domain signal of each frame to obtain a time-frequency signal; calculating the average value of the time-frequency signal of a preset frequency band to obtain a short-time frequency band average energy; calculating the variance of the short-time frequency band average energy to obtain a frequency band average energy variance.

[0008] In one embodiment, identifying the noise signal as a time-domain discontinuous abnormal sound signal based on the frequency band average energy variance includes: comparing the frequency band average energy variance with a first threshold, and when the frequency band average energy variance is greater than the first threshold, identifying the noise signal as a time-domain discontinuous abnormal sound signal.

[0009] In one embodiment, obtaining the first average observation value of the noise signal includes: obtaining a frame rate energy value based on the time domain signal of the fan; obtaining the Mel frequency coefficient of each frame of the time domain signal based on the frame rate energy value; predicting the Mel frequency coefficient of each frame based on a first mathematical model to obtain the first observation value of each frame signal in the time domain signal; removing the maximum frame and the minimum frame in the first observation value of each frame signal, and calculating the average of the first observation values ​​of the remaining frames to obtain the first average observation value.

[0010] In one embodiment, identifying the noise signal as an abnormal signal with a larger background noise based on the first average observation value includes: comparing the first average observation value with a second threshold, and identifying the noise signal as an abnormal signal with a larger background noise when the first average observation value is greater than the second threshold.

[0011] In one embodiment, obtaining the second average observation value of the noise signal includes: obtaining a frame rate energy value based on the time domain signal of the fan; obtaining the Mel frequency coefficient of each frame of the time domain signal based on the frame rate energy value; predicting the Mel frequency coefficient of each frame based on a second mathematical model to obtain the second observation value of each frame signal in the time domain signal; removing the maximum frame and the minimum frame in the second observation value of each frame signal, and calculating the average of the second observation values ​​of the remaining frames to obtain a second average observation value.

[0012] In one embodiment, identifying the noise signal as a time-domain continuous abnormal sound signal based on the second average observation value includes: comparing the second average observation value with a third threshold, and identifying the noise signal as a time-domain continuous abnormal sound signal when the second average observation value is greater than the third threshold.

[0013] The present invention also provides an abnormal sound detection device, which includes: a processor and a memory; an abnormal sound detection control program is stored on the processor, and the abnormal sound detection control program implements the steps of the abnormal sound detection method when executed by the processor.

[0014] The present invention proposes an abnormal noise detection method and abnormal noise detection device, which are used to detect the noise of the fan of the product. The abnormal noise detection method includes: collecting the noise signal generated when the fan is working; sequentially identifying the abnormal noise signal of time-domain discontinuous, the abnormal signal of background noise is too large, and the abnormal noise signal of time-domain continuous, and generating the identification result; according to the identification result, determining whether the fan has an abnormal noise problem. This patent collects the fan noise signal, and sequentially identifies the abnormal noise signal of time-domain discontinuous, the abnormal signal of background noise is too large, and the abnormal noise signal of time-domain continuous, and generates the identification result; according to the identification result, determining whether the fan has an abnormal noise problem according to the identification result, making the product quality control more refined and improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0016] Figure 1 A method step diagram of an abnormal sound detection method of the present invention;

[0017] Figure 2 A method step diagram of another embodiment of an abnormal sound detection method of the present invention;

[0018] Figure 3 A method step diagram of another embodiment of an abnormal sound detection method of the present invention;

[0019] Figure 4 A method step diagram of another embodiment of an abnormal sound detection method of the present invention;

[0020] Figure 5 A method step diagram of another embodiment of an abnormal sound detection method of the present invention;

[0021] Figure 6 This is a method step diagram of another embodiment of an abnormal sound detection method of the present invention.

[0022] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0023] 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.

[0024] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). When the specific posture changes, the directional indication also changes accordingly.

[0025] In the present invention, unless otherwise clearly specified and limited, the terms "connection", "fixation", etc. should be understood in a broad sense. For example, "fixation" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0026] In addition, in the present invention, descriptions such as "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0027] It is understandable that fan noise is an important quality indicator. For products such as computers, servers, game consoles and other electronic products, the abnormal noise of the fan will seriously affect the user experience, and may also mean manufacturing quality problems such as loose parts and uneven wear. Therefore, it is crucial for the designers and manufacturers of consumer electronic products with fans to promptly identify and solve the problem of abnormal fan noise. The abnormal noise emitted by the product fan has randomness in the time domain and frequency domain, and the abnormal fan noise is often generated by multiple sources. In addition to the turbulent noise generated by the fan itself, there may also be abnormal noises generated by the abnormal axis and fan blades of the fan itself, noises caused by the resonance of the fan and mechanical components, interference noises generated by foreign objects and fans, and noises from other sources. This makes it difficult to effectively capture the characteristics of abnormal fan noises through traditional time-frequency analysis methods. At present, there are some methods on the market for identifying abnormal fan noises, such as interception using spectrum analysis, statistical characteristics analysis or artificial listening methods. Due to the strong randomness of fan noise, poor spectrum consistency, and a wide variety of abnormal noises, the use of spectrum recognition methods has problems with high misjudgment and missed judgment rates. And once the test equipment has an abnormality and lacks self-detection capability, it is easy to cause the fan noise recognition of a large number of products to fail. The use of manual listening introduces subjectivity, resulting in inconsistent and unreliable recognition results. In addition, there are problems such as high labor costs, low efficiency, easy fatigue and high error rate.

[0028] Therefore, in order to more comprehensively and accurately identify abnormal fan noise of a product, the present invention proposes an abnormal noise detection method for detecting the noise of the fan of the product. Figure 1 , the method comprising:

[0029] S100, collecting a noise signal generated when the fan is working;

[0030] S200, sequentially performing time-domain discontinuous abnormal sound signal recognition, time-domain large abnormal sound signal recognition, and time-domain continuous abnormal sound signal recognition on the noise signal, and generating a recognition result;

[0031] S300: Determine whether the fan has an abnormal sound problem according to the identification result.

[0032] It is understandable that there are some methods on the market for identifying and solving the problem of abnormal fan noise, such as intercepting abnormal noise signals by using spectrum analysis, statistical characteristics analysis or artificial listening. However, fan noise is highly random, the spectrum consistency is poor, and there are many types of abnormal noise. The methods using spectrum analysis or statistical characteristics analysis have the problems of high misjudgment rate and missed judgment rate, and the methods using artificial listening have the problems of high error rate and cost. In order to be able to more comprehensively and accurately identify the abnormal fan noise of the product, this patent classifies the abnormal noise generated by complex and multi-source fans and other abnormal reasons of the product into time domain discontinuous abnormal noise, excessive background noise and time domain continuous abnormal noise. By collecting fan noise signals and performing time domain discontinuous abnormal noise recognition, abnormal signal judgment of excessive background noise and time domain continuous abnormal noise recognition on the noise signals generated by the fan, it is determined whether the fan has an abnormal noise problem based on the recognition results, making product quality control more refined and improving the user experience.

[0033] In this embodiment, before performing noise detection on the fan of the product, the technicians use different methods to systematically identify the various complex and changeable adverse phenomena in the product production test by extracting the signal differences of the product. Specifically, for the abnormal fan noise of the product, the technicians divide the abnormal noise of the product fan into three types according to the characteristics of the abnormal noise of the product fan, and formulate a signal interception model according to the characteristics of different categories. Through the formulated signal interception model, the abnormal noise of the product fan can be identified more comprehensively and accurately. Specifically, the noise signal generated when the fan is working is collected, and three types of abnormal sound detection are performed on the noise signal, which are to determine whether the noise signal of the fan is a non-continuous abnormal sound in the time domain, to determine whether the noise signal of the fan is an abnormal signal with a larger background noise, and to determine whether the noise signal of the fan is a continuous abnormal sound signal in the time domain. Among them, the abnormal noise that is non-continuous in the time domain means that the abnormal noise has no fixed pattern, and the time and interval of the abnormal noise are unpredictable. It suddenly appears at certain specific time points, may disappear for a period of time, and then appear again, presenting an intermittent state. By detecting whether the noise signal of the fan is an abnormal noise that is non-continuous in the time domain, it can be determined whether the noise signal of the fan has no fixed pattern; the background noise is too large, which means that the noise level in the environment exceeds its normal range, which may have an adverse effect on the detection and processing of various fan signals. By detecting whether the noise signal of the fan is an abnormal signal with a large background noise, it can be determined whether the noise signal of the fan exceeds the standard background noise range in the environment where the fan is located; the abnormal noise that is continuous in the time domain means that the abnormal noise has a fixed pattern, and the time and interval of the abnormal noise are predictable and continue to exist in time. By detecting whether the noise signal of the fan is an abnormal noise signal that is continuous in the time domain, it can be determined whether the noise signal of the fan has a fixed pattern. The noise signal is characterized by the above three detection methods to obtain three analysis results, and the signal type of the noise signal is determined according to the three analysis results, for example, whether the noise signal is a time-domain discontinuous abnormal sound signal, whether the noise signal is an abnormal signal with a relatively large background noise, and whether the noise signal is a time-domain continuous abnormal sound signal. When the noise signal is any of the three signal types of a time-domain discontinuous abnormal sound signal, a time-domain large abnormal sound signal, or a time-domain continuous abnormal sound signal, it is determined whether the fan has an abnormal sound problem.

[0034] In practical applications, the noise signal of the fan is obtained, and the noise signal of the fan is identified as a non-continuous abnormal noise signal in the time domain, an abnormal signal with a relatively large background noise, and an abnormal noise signal with a continuous time domain. When the signal interception model intercepts the abnormal signal of non-continuous abnormal noise in the time domain, the noise signal is identified as an abnormal signal of non-continuous abnormal noise in the time domain according to the interception action of the signal interception model, and it is determined that the current fan has an abnormal noise problem; or, when the signal interception model intercepts the abnormal signal of a relatively large background noise, the noise signal is identified as an abnormal signal with a relatively large background noise according to the interception action of the signal interception model, and it is determined that the current fan has an abnormal noise problem; or, when the signal interception model intercepts the abnormal signal of continuous abnormal noise in the time domain, the noise signal is identified as a continuous abnormal noise signal in the time domain according to the interception action of the signal interception model, and it is determined that the current fan has an abnormal noise problem. Therefore, according to the signal type of the signal intercepted by the signal interception model, the noise signal can be identified as an abnormal noise signal, and whether the fan has an abnormal noise problem can be determined according to the identification result, so as to more comprehensively and accurately identify the abnormal fan noise of the product.

[0035] It should be understood that the detection of abnormal fan noise of products can be carried out both during the production process of the products and after the products leave the factory. For example, testing during the production process can timely detect product abnormalities, and can detect problems such as abnormal fan noise before the products are fully assembled or leave the factory, so as to facilitate timely adjustments and repairs, avoid defective products from flowing into the next production link, and reduce subsequent rework costs. It helps to ensure that the products shipped have high quality standards in terms of fan operation and improve the reliability and stability of the products. Testing after the products leave the factory can be closer to the actual use scenario, taking into account different working conditions and load conditions, and a more comprehensive evaluation of the fan's performance. In addition, testing after leaving the factory can determine the root cause of the problem so that effective solutions can be taken to improve customer satisfaction. Among them, the products can be VR devices, laptops, servers, game consoles and other electronic products, which are usually equipped with cooling fans. Products can also be industrial equipment, such as large equipment such as machine tools and automated production lines, in which the cooling system and fans also need to ensure their normal operation during the testing process. The product may also be a household appliance, such as an air conditioner, a refrigerator, etc. The fans in the above appliances play a vital role in the normal operation of the equipment. When abnormal sound detection is implemented through the abnormal sound detection method, the fans in the above appliances are the focus of attention.

[0036] The present invention proposes an abnormal noise detection method for detecting noise on a fan of a product, the method comprising: collecting noise signals generated when the fan is working; sequentially identifying abnormal noise signals with discontinuous time domain, abnormal signals with excessive background noise, and abnormal noise signals with continuous time domain, and generating identification results; and determining whether the fan has an abnormal noise problem based on the identification results. This patent classifies abnormal noises generated by complex and multi-source fans and other abnormal causes of products into discontinuous abnormal noises in time domain, excessive background noise, and continuous abnormal noises in time domain. By collecting fan noise signals and sequentially identifying abnormal noises with discontinuous time domain, abnormal signals with excessive background noise, and continuous abnormal noises in time domain, it is determined whether the fan has an abnormal noise problem based on the identification results, making product quality control more refined and improving the user experience.

[0037] In one embodiment, referring to Figure 2 , the noise signal is sequentially identified as a non-continuous abnormal sound signal in the time domain, an abnormal signal with a large background noise, and an abnormal sound signal with a continuous time domain, and the generation of the identification result includes:

[0038] S210, obtaining a frequency band average energy variance of the noise signal, and identifying the noise signal as a time-domain discontinuous abnormal sound signal according to the frequency band average energy variance;

[0039] S220, obtaining a first average observation value of the noise signal, and identifying the noise signal as an abnormal signal with a larger background noise according to the first average observation value;

[0040] S230. Obtain a second average observation value of the noise signal, and identify the noise signal as a time-domain continuous abnormal sound signal according to the second average observation value.

[0041] It can be understood that, in this embodiment, the frequency band average energy variance D of the noise signal is obtained, and the frequency band average energy variance D can reflect the stability of the signal energy of the noise signal in the frequency band in time. For example, when the frequency band average energy variance D is small, it means that the energy of the noise signal in each frequency band is relatively stable and the noise signal is stable; when the frequency band average energy variance D is large, it means that the energy of the noise signal in each frequency band will change greatly, resulting in an increase in the frequency band average energy variance and the noise signal is not stable. Therefore, by analyzing the frequency band average energy variance D of the noise signal, it can be effectively determined whether the noise signal is a non-continuous abnormal sound signal in the time domain. The first average observation value V1 of the noise signal is obtained, and the evaluation score of the noise signal can be reflected by the first average observation value V1. For example, a relatively large first average observation value V1 indicates a noisy sound; a relatively small first average observation value V1 indicates a soft sound. Therefore, by analyzing the first average observation value of the noise signal, it can be effectively determined whether the noise signal is an abnormal signal with a large background noise. The second average observation value V2 of the noise signal is obtained, and the stability of the noise signal can be reflected by the second average observation value V2. For example, if the second average observation value V2 fluctuates slightly, it means that the noise signal is relatively stable; if the second average observation value V2 fluctuates greatly, it means that the noise signal is unstable. Therefore, by analyzing the second average observation value of the noise signal, it can provide an important basis for judging whether the noise signal is a continuous abnormal sound signal in the time domain.

[0042] In one embodiment, referring to Figure 3 , determining whether the fan has an abnormal sound problem according to the identification result includes:

[0043] S310, when the noise signal is identified as any one of an abnormal sound signal that is discontinuous in the time domain, an abnormal signal with a large background noise, and an abnormal sound signal that is continuous in the time domain, it is determined that the fan has an abnormal sound problem;

[0044] S320: When it is identified that the noise signals are not abnormal noise signals that are discontinuous in the time domain, abnormal noise signals with large background noise, and abnormal noise signals that are continuous in the time domain, it is determined that the fan does not have an abnormal noise problem.

[0045] It can be understood that in this embodiment, as long as the noise signal is identified as any one of a time-domain discontinuous abnormal sound signal, an abnormal signal with a relatively large background noise, and a time-domain continuous abnormal sound signal, the noise signal can be considered to be an abnormal signal, and it is determined that the fan has an abnormal sound problem. For example, when the noise signal is a time-domain discontinuous abnormal sound signal, or when the noise signal is both a time-domain discontinuous abnormal sound signal and an abnormal signal with a relatively large background noise, it is determined that the fan has an abnormal sound problem. In addition, only when the noise signal is not a time-domain discontinuous abnormal sound signal, an abnormal signal with a relatively large background noise, and a time-domain continuous abnormal sound signal, can it be determined that the fan does not have an abnormal sound problem.

[0046] In one embodiment, referring to Figure 4 , the obtaining of the frequency band average energy variance of the noise signal comprises:

[0047] S211, collecting a noise signal of the fan, and acquiring a time domain signal according to the noise signal;

[0048] S212, performing high-pass filtering on the time domain signal to obtain a filtered time domain signal;

[0049] S213, acquiring time domain signals of multiple frames according to the filtered time domain signal;

[0050] S214, performing Fourier transform processing on the time domain signal of each frame to obtain a time-frequency signal;

[0051] S215, calculating the average value of the time-frequency signal of the preset frequency band to obtain the short-time frequency band average energy;

[0052] S216: Calculate the variance of the short-time frequency band average energy to obtain the frequency band average energy variance.

[0053] It can be understood that in this embodiment, the noise signal generated by the fan when it is working is collected, and the time domain signal X(t) is obtained according to the noise signal. The t in the time domain signal X(t) represents time. The time domain signal X(t) is used to describe the change of the noise signal with time as a variable, indicating the specific value of the noise signal at different times. The time domain signal X(t) is subjected to high-pass filtering to obtain the filtered time domain signal . For the filtered time domain signal Perform frame processing to convert the time domain signal Split the signal into several time periods to obtain time domain signals of multiple frames , i represents the frame number or other variable number related to the frame. Perform Fourier transform (FFT) on each frame signal to obtain the time-frequency signal S(t,f), specifically S(t,f)= , where t in S(t,f) represents time and f represents frequency. The time-frequency signal S(t,f) is used to describe the manifestation of the noise signal from both time and frequency dimensions. The time-frequency signal S(t,f) can be used to obtain the temporal changes in each frequency component of the noise signal. Select the frequency band where the non-continuous abnormal sound energy of the fan noise is concentrated in the time domain. For example, when the f1-f2 frequency band is the frequency band where the non-continuous abnormal sound energy of the fan noise is concentrated in the time domain, set f1-f2 as the preset frequency band to filter and extract the signal, extract the most valuable information from the complex noise signal, and thus filter out the average intensity level of the noise signal in a specific frequency band. According to the preset frequency band f1-f2, sum all sample values ​​of the time-frequency signal S(t,f) and divide them by the total number of samples to obtain the short-time frequency band average energy E(t). The formula is: . The short-time frequency band average energy E(t) is used to indicate the average energy of the noise signal in different frequency bands within a short time t, that is, the short-time frequency band average energy E(t) reflects the distribution of the signal energy in each frequency range and the overall intensity level within a short time t. Calculate the variance of the short-time frequency band average energy E(t) to obtain the frequency band average energy variance D. Specifically, first obtain the time domain variance, remove the maximum and minimum values ​​in the time domain variance, reduce the impact of outliers on the results, and sum the time domain variance after removing the extreme values, and then divide it by the number of remaining time domain variances to finally obtain the frequency band average energy variance D. The formula is By analyzing the frequency band average energy variance D of the noise signal, it is possible to effectively determine whether the noise signal is a non-continuous abnormal sound signal in the time domain.

[0054] It should be understood that if there are multiple types of time-domain discontinuous abnormal noises in the fan noise time domain, or to avoid accidental abnormal interference. In other embodiments, multiple frequency bands in the fan noise time domain can be selected, such as f3-f4, f5-f6, etc., and the short-time frequency band average energy E(t) can be calculated according to the multiple frequency bands respectively to calculate the frequency band average energy variance D of the multiple frequency bands, so as to more comprehensively and accurately identify whether the noise signal is a time-domain discontinuous abnormal noise signal.

[0055] In one embodiment, referring to Figure 4 The step of identifying the noise signal as a time-domain discontinuous abnormal sound signal according to the frequency band average energy variance comprises:

[0056] S217: Compare the frequency band average energy variance with a first threshold value, and when the frequency band average energy variance is greater than the first threshold value, identify the noise signal as a time-domain discontinuous abnormal sound signal.

[0057] It can be understood that in this embodiment, the frequency band average energy variance D can reflect the temporal stability of the signal energy of the noise signal within the frequency band. When the frequency band average energy variance D is small, the noise signal is stable; when the frequency band average energy variance D is large, the noise signal fluctuates greatly. By analyzing the frequency band average energy variance D of the noise signal, it is possible to effectively determine whether the noise signal is a time-domain discontinuous abnormal sound signal. According to the characteristics of the frequency band average energy variance D, this embodiment is provided with a first threshold value, and the frequency band average energy variance D is compared with the first threshold value. When the frequency band average energy variance D is greater than the first threshold value, the signal interception model will intercept the noise signal as a time-domain discontinuous abnormal sound signal, and at this time, the noise signal is identified as a time-domain discontinuous abnormal sound signal; when the frequency band average energy variance D is less than or equal to the first threshold value, the signal interception model will not perform signal interception, and at this time, it is identified that the noise signal is not a time-domain discontinuous abnormal sound signal.

[0058] In one embodiment, referring to Figure 5 , the obtaining of the first average observation value of the noise signal comprises:

[0059] S221, obtaining a frame rate energy value according to the time domain signal of the fan;

[0060] S222, obtaining a Mel frequency coefficient of each frame of the time domain signal according to the frame rate energy value;

[0061] S223, predicting the Mel frequency coefficient of each frame based on the first mathematical model to obtain a first observation value of each frame signal in the time domain signal;

[0062] S224. Remove the maximum frame and the minimum frame from the first observation values ​​of each frame of the signal, and calculate the average of the first observation values ​​of the remaining frames to obtain a first average observation value.

[0063] It can be understood that, in this embodiment, the time domain signals of multiple frames are obtained. , according to the preset frame length, multiple frames of time domain signals are Frame the time domain signal Divide into multiple signal frames, calculate the energy of each frame signal, and calculate the sum of the squares of the amplitude of each frame signal to obtain the frame rate energy value X(i,k). The formula is X(i,k)= . The i in the frame rate energy value X(i,k) represents the frame number or other variable number related to the frame, and the k in the frame rate energy value X(i,k) represents the energy value. The frame rate energy value X(i,k) is used to reflect the energy size of the noise signal in a certain frame. MFCC feature extraction is performed based on the frame rate energy value X(i,k), and MFCC feature extraction is performed on the frame rate energy value X(i,k) through the Mel frequency cepstrum coefficient (MFCC), so as to obtain the Mel frequency coefficient M(i,m) of each frame. The i in the Mel frequency coefficient M(i,m) of each frame represents the frame number or other variable number related to the frame, and the m in the Mel frequency coefficient M(i,m) of each frame represents the Mel frequency band. The formula is M(i,m)= The Mel frequency coefficient M(i,m) of each frame is predicted using the pre-trained first mathematical model SVM1 to obtain the first observation value of each frame signal in the time domain signal. , remove the first observation to remove the first observation of each frame signal The maximum 10% frames and the minimum 10% frames are selected, and the mean of the observation values ​​of the remaining frames is calculated to finally obtain the first average observation value V1, thereby making the first average observation value V1 calculated finally more stable. Among them, the first mathematical model SVM1 is implemented by the nonlinear support vector machine model SVM1. The technicians train the nonlinear support vector machine model SVM1 in advance by accessing sample data, preprocessing data and extracting MFCC features.

[0064] It should be understood that before MFCC feature extraction, the number of abnormal frames needs to be calculated to ensure the accuracy of MFCC feature extraction. Therefore, in this embodiment, an energy threshold is also provided to check the energy value of each frame in the frame rate energy value X(i, k). When it is lower than or higher than the energy threshold, it is judged as an abnormal frame, the number of abnormal frames is counted, and the abnormal frames are excluded, thereby improving the accuracy of MFCC feature extraction. At the same time, the number of abnormal frames is output to the server as a result for users to evaluate signal quality, adjust parameters, etc.

[0065] In one embodiment, referring to Figure 5 The step of identifying the noise signal as an abnormal signal with a large background noise according to the first average observation value includes:

[0066] S225. Compare the first average observation value with a second threshold value, and when the first average observation value is greater than the second threshold value, identify the noise signal as an abnormal signal with a relatively large background noise.

[0067] It can be understood that, in this embodiment, the first average observation value V1 can reflect the evaluation score of the noise signal. For example, a relatively large first average observation value V1 indicates a noisy sound; a relatively small first average observation value V1 indicates a soft sound. Therefore, by analyzing the first average observation value of the noise signal, it is possible to effectively determine whether the noise signal is an abnormal signal with a relatively large background noise. In this embodiment, a second threshold is set according to the characteristics of the first average observation value V1, and the first average observation value V1 is compared with the second threshold. When the first average observation value V1 is greater than the second threshold, the signal interception model will intercept the noise signal as an abnormal signal with a relatively large background noise, and at this time, the noise signal is identified as an abnormal signal with a relatively large background noise; when the first average observation value V1 is less than or equal to the second threshold, the signal interception model will not perform signal interception, and at this time, the noise signal is identified as not being an abnormal signal with a relatively large background noise.

[0068] In one embodiment, referring to Figure 6 , the obtaining of the second average observation value of the noise signal comprises:

[0069] S231, obtaining a frame rate energy value according to the time domain signal of the fan;

[0070] S232, obtaining a Mel frequency coefficient of each frame of the time domain signal according to the frame rate energy value;

[0071] S233, predicting the Mel frequency coefficient of each frame based on the second mathematical model to obtain a second observation value of each frame signal in the time domain signal;

[0072] S234. Remove the maximum frame and the minimum frame from the second observation values ​​of each frame of the signal, and calculate the average value of the second observation values ​​of the remaining frames to obtain a second average observation value.

[0073] It can be understood that, in this embodiment, the time domain signals of multiple frames are obtained. , according to the preset frame length, multiple frames of time domain signals are Frame the time domain signal Divide into multiple signal frames, calculate the energy of each frame signal, and calculate the sum of the squares of the amplitude of each frame signal to obtain the frame rate energy value X(i,k). The i in the frame rate energy value X(i,k) represents the frame number or other variable number related to the frame, and the k in the frame rate energy value X(i,k) represents the energy value. The frame rate energy value X(i,k) is used to reflect the energy size of the noise signal in a certain frame. MFCC feature extraction is performed based on the frame rate energy value X(i,k), and the MFCC feature extraction is performed on the frame rate energy value X(i,k) through the Mel frequency cepstrum coefficient (MFCC), so as to obtain the Mel frequency coefficient M(i,m) of each frame. The i in the Mel frequency coefficient M(i,m) of each frame represents the frame number or other variable number related to the frame, and the m in the Mel frequency coefficient M(i,m) of each frame represents the Mel frequency band. The formula is M(i,m)= The Mel frequency coefficient M(i,m) of each frame is predicted using the pre-trained second mathematical model SVM2 to obtain the second observation value of each frame signal in the time domain signal. , remove the second observation to remove the second observation value of each frame signal The maximum 10% frames and the minimum 10% frames in the image are selected, and the mean of the observation values ​​of the remaining frames is calculated to finally obtain the second observation value V2, thereby making the second average observation value V2 calculated in the end more stable. Among them, the second mathematical model SVM2 is implemented by the nonlinear support vector machine model SVM2. The technicians train the nonlinear support vector machine model SVM2 in advance by accessing sample data, preprocessing data and extracting MFCC features.

[0074] In one embodiment, referring to Figure 6 , the step of identifying the noise signal as a time-domain continuous abnormal sound signal according to the second average observation value comprises:

[0075] S235. Compare the second average observation value with a third threshold value, and when the second average observation value is greater than the third threshold value, identify the noise signal as a time-domain continuous abnormal sound signal.

[0076] It can be understood that in this embodiment, the stability of the noise signal can be reflected by the second average observation value V2. For example, the second average observation value V2 fluctuates slightly, indicating that the noise signal is relatively stable; the second average observation value V2 fluctuates greatly, indicating that the noise signal is unstable. Therefore, by analyzing the second average observation value of the noise signal, an important basis can be provided for judging whether the noise signal is an abnormal sound signal that is continuous in the time domain. In this embodiment, according to the characteristics of the second average observation value V2, a third threshold is set, and the second average observation value V2 is compared with the third threshold. When the second average observation value V2 is greater than the third threshold, the signal interception model will intercept the noise signal as an abnormal sound signal that is continuous in the time domain, and the noise signal is identified as a continuous abnormal sound signal in the time domain; when the second average observation value V2 is less than or equal to the third threshold, the signal interception model will not perform signal interception, and the noise signal is identified as not being a continuous abnormal sound signal in the time domain.

[0077] The present invention further provides an abnormal sound detection device, comprising: a processor and a memory; an abnormal sound detection control program is stored on the processor, and when the abnormal sound detection control program is executed by the processor, the steps of the abnormal sound detection method described in the above embodiment are implemented.

[0078] It can be understood that in order to be able to more comprehensively and accurately identify the abnormal fan noise of the product, the present invention proposes an abnormal sound detection device, which classifies the abnormal noise generated by the complex and multi-source fans and other abnormal causes of the product into time-domain discontinuous abnormal noise, excessive background noise and time-domain continuous abnormal noise, and performs time-domain discontinuous abnormal sound recognition, excessive background noise abnormal signal judgment and time-domain continuous abnormal sound recognition on the noise signal generated by the fan in turn, and can determine whether to perform an interception action on the abnormal sound signal through a signal interception model, thereby identifying the signal type of the noise signal according to the interception action, and finally determining whether the fan has an abnormal noise problem according to the identification result, thereby making product quality control more refined and improving the user experience.

[0079] The above descriptions are only optional embodiments of the present invention, and are not intended to limit the patent scope of the present invention. All equivalent structural changes made using the contents of the present invention's specification and drawings, or directly / indirectly applied in other related technical fields, are included in the patent protection scope of the present invention.

Claims

1. A method for detecting abnormal sound, characterized in that: For detecting the noise of a fan of a product, the method comprising: Collecting the noise signal generated when the fan is working; The noise signal is sequentially subjected to time-domain discontinuous abnormal sound signal recognition, time-domain large abnormal signal recognition, and time-domain continuous abnormal sound signal recognition, and a recognition result is generated; Determining whether the fan has an abnormal sound problem according to the identification result; The step of sequentially performing time-domain discontinuous abnormal sound signal recognition, time-domain abnormal signal recognition with large background noise, and time-domain continuous abnormal sound signal recognition on the noise signal and generating recognition results includes: Obtaining a frequency band average energy variance of the noise signal, and identifying the noise signal as a time-domain discontinuous abnormal sound signal according to the frequency band average energy variance; Obtaining a first average observation value of the noise signal, and identifying the noise signal as an abnormal signal with a larger background noise according to the first average observation value; Acquire a second average observation value of the noise signal, and identify the noise signal as a time-domain continuous abnormal sound signal according to the second average observation value; The obtaining of the first average observation value of the noise signal comprises: Acquire a frame rate energy value according to a time domain signal of the fan; Acquire the Mel frequency coefficient of each frame of the time domain signal according to the frame rate energy value; Based on the first mathematical model, predicting the Mel frequency coefficient of each frame to obtain a first observation value of each frame signal in the time domain signal; Removing the maximum frame and the minimum frame from the first observation values ​​of each frame of the signal, and calculating the average of the first observation values ​​of the remaining frames to obtain a first average observation value; The obtaining of the second average observation value of the noise signal comprises: Acquire a frame rate energy value according to a time domain signal of the fan; Acquire the Mel frequency coefficient of each frame of the time domain signal according to the frame rate energy value; Based on the second mathematical model, predict the Mel frequency coefficient of each frame to obtain a second observation value of each frame signal in the time domain signal; The maximum frame and the minimum frame in the second observation value of each frame signal are removed, and the average value of the second observation values ​​of the remaining frames is calculated to obtain a second average observation value.

2. The abnormal sound detection method according to claim 1, characterized in that: Determining whether the fan has an abnormal sound problem according to the identification result includes: When the noise signal is identified as any one of an abnormal sound signal that is discontinuous in the time domain, an abnormal signal with a large background noise, and an abnormal sound signal that is continuous in the time domain, it is determined that the fan has an abnormal sound problem; When it is identified that the noise signals are not abnormal noise signals that are discontinuous in the time domain, abnormal signals with relatively large background noise, and abnormal noise signals that are continuous in the time domain, it is determined that the fan does not have an abnormal noise problem.

3. The abnormal sound detection method according to claim 1, characterized in that: The step of obtaining the frequency band average energy variance of the noise signal comprises: Collecting a noise signal of the fan, and acquiring a time domain signal according to the noise signal; Performing high-pass filtering on the time domain signal to obtain a filtered time domain signal; Acquire time domain signals of multiple frames according to the filtered time domain signals; Perform Fourier transform processing on the time domain signal of each frame to obtain the time-frequency signal; Calculate the average value of the time-frequency signal of the preset frequency band to obtain the average energy of the short-time frequency band; The variance of the short-time frequency band average energy is calculated to obtain the frequency band average energy variance.

4. The abnormal sound detection method according to claim 1, characterized in that: The step of identifying the noise signal as a time-domain discontinuous abnormal sound signal according to the frequency band average energy variance comprises: The frequency band average energy variance is compared with a first threshold, and when the frequency band average energy variance is greater than the first threshold, the noise signal is identified as a time-domain discontinuous abnormal sound signal.

5. The abnormal sound detection method according to claim 1, characterized in that: The step of identifying the noise signal as an abnormal signal with a large background noise according to the first average observation value comprises: The first average observation value is compared with a second threshold value, and when the first average observation value is greater than the second threshold value, the noise signal is identified as an abnormal signal with a relatively large background noise.

6. The abnormal sound detection method according to claim 1, characterized in that: The step of identifying the noise signal as a time-domain continuous abnormal sound signal according to the second average observation value comprises: The second average observation value is compared with a third threshold value, and when the second average observation value is greater than the third threshold value, the noise signal is identified as a time-domain continuous abnormal sound signal.

7. An abnormal sound detection device, characterized in that: The abnormal sound detection device comprises: Processor and memory; An abnormal sound detection control program is stored on the processor, and when the abnormal sound detection control program is executed by the processor, the steps of the abnormal sound detection method according to any one of claims 1 to 6 are implemented.

Images