Automatic classification method and system based on machine sound time characteristics

Abstract

The invention discloses an automatic classification method and system based on machine sound time characteristics, and the method comprises the steps: 1) carrying out the feature extraction of an original sound signal, and obtaining a logarithm Mel spectrum spectrogram (feature matrix); 2) calculating a similarity coefficient of each frame in the logarithm Mel spectrogram by using a method based on feature similarity measurement; 3) calculating the similarity coefficient variance of the spectrogram according to the similarity coefficient in the step 2); 4) processing all the sound signals of each machine ID according to the steps 1)-3) to obtain similarity coefficient variances of all the sound signals in the machine ID; 5) comparing each similarity coefficient variance in the step 4) with an experience variance, and deciding the stability of each sound signal; and 6) counting the specific gravity of the stationary signal and the non-stationary signal, and comparing the specific gravity with a set threshold so as to judge the overall stability of each machine ID. According to the method, the stationarity and non-stationarity classification task of the complex industrial sound signals can be quickly and efficiently completed, and accurate signal classification is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal classification, and in particular relates to an automatic classification method and system based on the time characteristics of machine sounds. Background Art

[0002] In the industrial production process, the abnormal state of the equipment is often accompanied by specific sound changes. Therefore, it has become an effective means to detect anomalies by analyzing the acoustic signals of the machine. The acoustic signal contains a lot of information related to the operating status of the equipment. For example, factors such as rotation speed and friction are reflected in the acoustic signal. By analyzing and processing the acoustic signal, potential failures of the equipment can be discovered in advance to prevent safety accidents. However, there are many types of industrial equipment, complex working modes, and the sound signals generated by different types of equipment often have significant differences. Generally, the acoustic signals emitted by industrial machines can be divided into two categories: stable sound signals and non-stationary sound signals. For example, when working, equipment such as fans usually generate sound signals with continuous and stable frequency characteristics. Such signals can be classified as stable sound signals; while valves may emit intermittent sound signals in different cycles. The sound characteristics vary due to periodic changes, so they are regarded as non-stationary sound signals. The distinction between stable and non-stationary signals is of great significance for abnormal sound signal detection. Especially in the automated abnormality detection system, identifying the stability of the signal helps to improve the detection accuracy and formulate corresponding classification standards for different types of equipment.

[0003] Most of the existing unsupervised abnormal sound detection methods extract the logarithmic Mel-spectrogram features of machine sounds and directly use all of their features for abnormal detection without considering the sound characteristics of the machine type, resulting in low accuracy in signal classification. In order to solve this problem, the present invention proposes an automatic classification method and system based on the time characteristics of machine sounds. Summary of the invention

[0004] In response to the above problems, the present invention proposes an automatic classification method and system based on the time characteristics of machine sound. The present invention first extracts the logarithmic Mel spectrum of the sound signal; then, calculates the similarity coefficient between each frame and the entire logarithmic Mel spectrum; then, calculates the variance of the similarity coefficient of the spectrum, and compares it with the empirical variance to determine the stationarity of each sound signal; finally, by counting the proportion of stationary and non-stationary sound signals and comparing them with the set threshold, the stationarity of each machine ID data set is determined to complete the automatic classification.

[0005] The following is a description of the professional terms involved in the present invention:

[0006] Log-Mel spectrogram is a commonly used audio signal processing method, which is used to convert audio signals into frequency domain representation. In the log-Mel spectrogram, one dimension on the time axis is called "time frame". When the target sound event occurs continuously over time, the related sound signal is called a stationary sound signal. When the target sound event rarely occurs, the related sound signal is called a non-stationary sound signal. The main information of the stationary sound signal is distributed throughout the time interval, and most of the spectrogram frames in its log-Mel spectrogram show the frequency characteristics of the target sound signal. Therefore, there is similarity between the spectrogram frames of the stationary sound signal. However, only a small number of spectrogram frames in the non-stationary sound signal are related to the target sound signal, which leads to the dissimilarity between the spectrogram frames containing the target sound and other spectrogram frames in the non-stationary sound signal.

[0007] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0008] The automatic classification method based on the time characteristics of machine sound calculates the similarity coefficient between each frame of the log-Mel Spectrogram of the sound signal and all other frames, and then obtains the similarity coefficient between each frame and the entire log-Mel Spectrogram. Then, based on the similarity coefficient of each frame, the variance of the similarity coefficient of the entire log-Mel Spectrogram is calculated, and the stationarity of each sound signal is judged by comparing the similarity coefficient variance with the empirical variance. Then, for each machine ID, the number of stationary sound signals and non-stationary sound signals in each ID data set is counted, and their proportions are calculated respectively. Finally, the proportion of stationary sound signals of each machine ID is compared with the set threshold to determine the stationarity of each machine ID and complete the automatic classification. The specific steps are as follows:

[0009] Step 1: extract the logarithmic Mel spectrum feature of the original sound signal, including pre-emphasis, framing, windowing, fast Fourier transform and logarithmic Mel filtering, to obtain its logarithmic Mel spectrum feature X.

[0010] Step 2: perform similarity calculation on the logarithmic Mel-spectrogram feature X obtained in step 1, by measuring the similarity of each frame x i With other frames x j (i≠j), and then obtain the similarity coefficient between each frame and the entire logarithmic Mel spectrum.

[0011] Step 3: For the similarity coefficient obtained in step 2, further calculate the variance of the similarity coefficient of the spectrum graph.

[0012] Step 4: Process the collected industrial equipment sound signal data set according to steps 1-3, and calculate the similarity coefficient variance of all sound signals in each machine ID data set.

[0013] Step 5, compare each similarity coefficient variance obtained in step 4 with the empirical variance to determine whether each sound signal is stationary or non-stationary.

[0014] Step 6: Based on the judgment results of all sound signals in each machine ID data set obtained in step 5, the number of signals judged to be stationary and non-stationary is counted, and the proportion is calculated. The calculated stationary ratio is compared with the set threshold θ to determine the overall stationarity of the machine ID.

[0015] Preferably, step 1 includes pre-emphasis, framing, windowing, fast Fourier transform and logarithmic Mel filtering, as follows:

[0016] Step 1.1, for pre-emphasis, the calculation formula is:

[0017] H(n)=1-αn -1

[0018] Where n represents the index of discrete time or sequence, H(n) is the output after pre-emphasis, and α is the pre-emphasis coefficient;

[0019] Step 1.2, for framing, the frame length is L and the frame shift is

[0020] Step 1.3, for windowing, use the Hamming window, and the calculation formula is:

[0021]

[0022] Step 1.4, for fast Fourier transform, convert the signal obtained in step 1.1-step 1.3 into the frequency domain, and the calculation formula is:

[0023]

[0024] Where e is the base of the natural logarithm, j is the imaginary unit, and j satisfies 2 = -1, y(n) is the windowed acoustic signal, Y(k) is the frequency domain signal corresponding to y(n), and N is the number of Fourier transform points;

[0025] Step 1.5: Process the spectrum Y(k) obtained in step 1.4 through a logarithmic Mel filter to obtain the final output logarithmic Mel spectrum feature X. The calculation formula is:

[0026] X = log(ω M ||FFT(y)|| 2 )

[0027] Among them, y=[y(0),y(1),...,y(N-1)], ω M ∈R d×Bis a Mel filter bank, R represents the real number domain, B represents each discrete frequency component of the spectrum obtained by FFT, and d represents the number of Mel filters.

[0028] Preferably, step 2 is as follows:

[0029] Step 2.1, after processing in step 1, a logarithmic Mel-spectrogram feature X is generated, which is defined as:

[0030] X∈R d×t =[x 1 ,x 2 ,...,x t ]

[0031] Among them, x i ∈R d×1 ,i∈[1,2,...,t], t represents the frame number;

[0032] Step 2.2, the similarity between each frame in the spectrogram and other frames is expressed as:

[0033] s ij =similarity(x i ,x j ),i≠j,(i,j∈[1,2,...,t])

[0034] Step 2.3, x i The similarity coefficient sim between the entire spectrum i It is expressed as:

[0035]

[0036] Preferably, in step 3, the similarity coefficient variance is calculated as:

[0037]

[0038] Among them, μ is the mean of the similarity coefficient, σ 2 is the variance of the similarity coefficient.

[0039] Preferably, in step 4, the obtained similarity coefficient variance is expressed as:

[0040]

[0041] Where m is the number of sound samples in the dataset for each machine ID.

[0042] Preferably, in step 5, after obtaining the similarity coefficient variance set in step 4, each similarity coefficient variance is compared with the empirical variance to determine the stationarity of each sound signal. The specific decision function is expressed as:

[0043]

[0044] Where δ is the empirical variance threshold.

[0045] Preferably, step 6 is as follows:

[0046] Step 6.1, based on the decision method in step 5, obtain the stationarity or non-stationarity determination results of all sound signals in the data set of each ID, which is specifically expressed as:

[0047] Category i ={S,N,S,S,N,....,S,N},i∈[00,01,02,03,04,05,06]

[0048] Among them, S indicates that the sound signal is judged to be a stationary sound signal, and N indicates that the sound signal is judged to be a non-stationary sound signal.

[0049] Step 6.2: For each machine ID, count the number of stationary sound signals and the number of non-stationary sound signals in its data set, denoted as N S and N N .

[0050] In step 6.3, considering the interference of noise and other redundant factors, there may be misjudgment of the stationary judgment of a single sound signal. Therefore, in order to improve the robustness of classification, based on step 6.2, the proportion of the number of stationary signals and non-stationary signals in the machine ID data set is calculated, which are defined as the stationary ratio (SR) and the non-stationary ratio (NR). Specifically expressed as:

[0051]

[0052] Among them, SR+NR=1.

[0053] Next, the calculated stability ratio (SR) is compared with the set threshold θ to determine the overall stability of each machine ID, which is specifically expressed as:

[0054]

[0055] The present invention also discloses an automatic classification system based on the time characteristics of machine sounds, which is used to execute the above method and includes the following modules:

[0056] Logarithmic Mel spectrum feature extraction module: extracts the logarithmic Mel spectrum feature of the original sound signal to obtain the logarithmic Mel spectrum feature X;

[0057] Similarity coefficient acquisition module: Calculate the similarity of the obtained logarithmic Mel spectrum feature X, by measuring each frame x i With other frames x j The similarity coefficient between each frame and the entire logarithmic Mel spectrum is obtained; where i≠j;

[0058] Similarity coefficient variance calculation module of the spectrum graph: for the obtained similarity coefficient, the similarity coefficient variance of the spectrum graph is calculated;

[0059] Similarity coefficient variance calculation module for all sound signals: processes the collected industrial equipment sound signal data set and calculates the similarity coefficient variance of all sound signals in each machine ID data set;

[0060] Sound signal stability judgment module: compares each similarity coefficient variance obtained with the empirical variance to determine whether each sound signal is stable;

[0061] Overall stability judgment module: Based on the judgment results of all sound signals in each machine ID data set, the number of signals judged as stable and non-stationary is counted, and the proportion is calculated. The calculated stability ratio is compared with the set threshold θ, and the overall stability of the machine ID is judged based on the comparison results.

[0062] The beneficial effects of the present invention are as follows:

[0063] The automatic classification method and system based on the time characteristics of machine sound of the present invention can quickly and efficiently complete the task of classifying the stationarity and non-stationarity of complex industrial sound signals, realize accurate signal classification, and significantly reduce the misjudgment caused by noise interference, thereby improving the robustness of the system.

[0064] Compared with the prior art, the present invention has faster classification speed, less memory usage, and greatly optimizes the efficiency of system resource utilization. The present invention has strong adaptability, can process various types of industrial equipment and complex audio data, performs well in automatic determination of stability, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a schematic diagram of a stable sound signal. Raw Audio is the original audio; Time is the time axis; Frequency is the frequency axis; Main information is the main information.

[0066] Figure 2 Schematic diagram of non-stationary signal.

[0067] Figure 3 This is a flow chart of an automatic classification method based on the time characteristics of machine sounds according to a preferred embodiment of the present invention.

[0068] Figure 4 This is a block diagram of an automatic classification system based on the time characteristics of machine sounds according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0069] The present invention is further described below using a data set in industrial equipment as an example, using an automatic classification method based on the time characteristics of machine sound. The following description is only for illustration and does not impose any form of limitation on the present invention.

[0070] In the present invention, the public data sets MIMII and ToyADMOS for industrial machine sound anomaly detection are used. The MIMII data set contains four types of industrial equipment: Fan, Pump, Slider, and Valve. Each equipment type has four machine IDs. The data set for each ID consists of approximately 1,000 normal sound signal samples for training and 100-200 normal and abnormal sound signal samples for testing. The sampling rate is 16KHz and the duration is 10s. The ToyADMOS data set contains two types of industrial equipment: ToyCar and ToyConveyor, with four and three machine IDs, respectively. The data set for each machine ID consists of approximately 1,000 normal sound signal samples for training and 200-400 normal and abnormal sound signal samples for testing. The sampling rate is 16KHz, and the duration is 11s and 10s, respectively.

[0071] like Figure 3 As shown, this embodiment is an automatic classification method based on the time characteristics of machine sounds, and the specific steps are as follows:

[0072] Step 1: extract the logarithmic Mel spectrum feature of the original sound signal, including pre-emphasis, framing, windowing, fast Fourier transform and logarithmic Mel filtering, and finally obtain the logarithmic Mel spectrum feature X. Specifically:

[0073] Step 1.1, for pre-emphasis, the calculation formula is:

[0074] H(n)=1-αn -1

[0075] Wherein, n is the index of discrete time or sequence, H(n) is the output after pre-emphasis, and α is the pre-emphasis coefficient, which is 0.97.

[0076] Step 1.2, for framing, the frame length is L and the frame shift is L is set to 64 milliseconds.

[0077] Step 1.3, for windowing, use the Hamming window, and the calculation formula is:

[0078]

[0079] Step 1.4, for fast Fourier transform, convert the signal obtained in step 1.1-step 1.3 into the frequency domain, and the calculation formula is:

[0080]

[0081] Where e is the base of the natural logarithm, j is the imaginary unit, and j satisfies 2 =-1, y(n) is the windowed acoustic signal, Y(k) is the frequency domain signal corresponding to y(n), and N is the number of Fourier transform points, which is 512 in this embodiment.

[0082] Step 1.5, the spectrum Y(k) obtained in step 1.4 is processed by a logarithmic Mel filter to obtain the final output X, which is calculated as follows:

[0083] X = log(ω M ||FFT(y)|| 2 )

[0084] Among them, y=[y(0),y(1),...,y(N-1)], ω M ∈R d×B is the Mel filter bank, R represents the real number domain, B is each discrete frequency component of the spectrum obtained by FFT, d represents the number of Mel filters, which is set to 128, the window size is set to 1024, and the frame shift is 512.

[0085] Step 2: Calculate the similarity feature value of the spectrogram frame for the logarithmic Mel-spectrogram feature X obtained in step 1.5, by measuring each frame x i With other frames x j (i≠j), and then obtain the similarity coefficient between each frame and the entire logarithmic Mel spectrum. Specifically:

[0086] Step 2.1, first, after processing in step 1, a logarithmic Mel spectrum matrix X is generated, which is defined as:

[0087] X∈R d×t =[x 1 ,x 2 ,...,x t ]

[0088] where x i ∈R d×1 ,i∈[1,2,...,t], where t represents the frame number;

[0089] Step 2.2, secondly, the two frames x in the spectrogram i and x j The similarity between them can be expressed as:

[0090] s ij =similarity(x i ,x j ),i≠j,(i,j∈[1,2,...,t])

[0091] Step 2.3, therefore, x i The similarity coefficient sim between the entire spectrum i It is expressed as:

[0092]

[0093] Step 3: The similarity coefficient sim obtained in step 2.3 i , and further calculate the similarity coefficient variance of the spectrum graph, expressed as:

[0094]

[0095] Among them, μ is the mean of the similarity coefficient, σ 2 is the variance of the similarity coefficient.

[0096] Step 4: Process the collected industrial equipment sound signal data set according to steps 1-3 to obtain the similarity coefficient variance of all sound signals in the data set. It is expressed as:

[0097]

[0098] Where m is the number of sound samples in the dataset for each machine ID.

[0099] Step 5: Set the similarity coefficient variance obtained in step 4 Compared with the empirical variance δ, the decision sound signal is stationary or non-stationary. The specific decision is expressed as:

[0100]

[0101] Where δ is the empirical variance threshold.

[0102] Step 6: Based on the judgment results of all sound signals in each machine ID data set obtained in step 5, the number of signals judged as stable and non-stable is counted, and the proportion is calculated. The calculated stability ratio is compared with the set threshold θ to judge the overall stability of the machine ID, where θ is 0.5. Specifically:

[0103] Step 6.1, based on the decision method in step 5, obtain the stationary or non-stationary determination results of all sound signals in the data set of each machine ID, which is specifically expressed as:

[0104] Category i={S,N,S,S,N,....,S,N},i∈[00,01,02,03,04,05,06]

[0105] Among them, S indicates that the sound signal is judged to be a stationary sound signal, and N indicates that the sound signal is judged to be a non-stationary sound signal.

[0106] Step 6.2: For each machine ID, count the number of stationary sound signals and the number of non-stationary sound signals in its data set, denoted as N S and N N .

[0107] Step 6.3, considering the interference of noise and other redundant factors, there may be misjudgment of the stationary judgment of a single sound signal. Therefore, in order to improve the robustness of classification, based on step 6.2, the proportion of the number of stationary signals and non-stationary signals in the machine ID data set is calculated, which are defined as stationary ratio (SR) and non-stationary ratio (NR). Specifically expressed as:

[0108]

[0109] Among them, SR+NR=1.

[0110] Next, the calculated stability ratio (SR) is compared with the set threshold θ to determine the overall stability of the machine ID, which is specifically expressed as:

[0111]

[0112] like Figure 4 As shown, this embodiment discloses an automatic classification system based on the time characteristics of machine sounds, which is used to execute the above method, and includes the following modules:

[0113] Logarithmic Mel spectrum feature extraction module: extracts the logarithmic Mel spectrum feature of the original sound signal to obtain the logarithmic Mel spectrum feature X;

[0114] Similarity coefficient acquisition module: Calculate the similarity of the obtained logarithmic Mel spectrum feature X, by measuring each frame x i With other frames x j The similarity coefficient between each frame and the entire logarithmic Mel spectrum is obtained; where i≠j;

[0115] Similarity coefficient variance calculation module of the spectrum graph: for the obtained similarity coefficient, the similarity coefficient variance of the spectrum graph is calculated;

[0116] Similarity coefficient variance calculation module for all sound signals: processes the collected industrial equipment sound signal data set and calculates the similarity coefficient variance of all sound signals in each machine ID data set;

[0117] Sound signal stability judgment module: compares each similarity coefficient variance obtained with the empirical variance to determine whether each sound signal is stable;

[0118] Overall stability judgment module: Based on the judgment results of all sound signals in each machine ID data set, the number of signals judged as stable and non-stationary is counted, and the proportion is calculated. The calculated stability ratio is compared with the set threshold θ, and the overall stability of the machine ID is judged based on the comparison results.

[0119] For other contents of this embodiment, please refer to the above method embodiment.

[0120] Any matters not described in the present invention are applicable to the prior art.

[0121] The specific implementation modes of the present invention are described above to facilitate the understanding of the present invention by those skilled in the art. However, the present invention is not limited to the scope of the specific implementation modes. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

Claims

1. An automatic classification method based on the time characteristics of machine sounds, characterized in that: The specific steps include: Step 1, extracting the logarithmic Mel spectrum feature of the original sound signal to obtain the logarithmic Mel spectrum feature X; Step 2: perform similarity calculation on the logarithmic Mel-spectrogram feature X obtained in step 1, by measuring the similarity of each frame x i With other frames x j The similarity coefficient between each frame and the entire logarithmic Mel spectrum is obtained; where i≠j; Step 3, for the similarity coefficient obtained in step 2, calculate the variance of the similarity coefficient of the spectrum graph; Step 4, process the collected industrial equipment sound signal data set according to steps 1-3, and calculate the similarity coefficient variance of all sound signals in each machine ID data set; Step 5, comparing each similarity coefficient variance obtained in step 4 with the empirical variance to determine whether each sound signal is stable; Step 6: Based on the judgment results of all sound signals in each machine ID data set obtained in step 5, the number of signals judged as stationary and non-stationary is counted, and the proportion is calculated. The calculated stationary ratio is compared with the set threshold θ, and the overall stationarity of the machine ID is judged based on the comparison result.

2. The automatic classification method based on the time characteristics of machine sounds as claimed in claim 1, characterized in that: Step 1 includes pre-emphasis, framing, windowing, fast Fourier transform and logarithmic Mel filtering, as follows: Step 1.1, for pre-emphasis, the calculation formula is H(n) = 1-αn -1 , where n is the index of discrete time or sequence, H(n) is the output after pre-emphasis, and α is the pre-emphasis coefficient; Step 1.2, for framing, the frame length is L and the frame shift is Step 1.3, for windowing, use the Hamming window, and the calculation formula is Step 1.4, for fast Fourier transform, convert the signal obtained in step 1.1-step 1.3 into the frequency domain, and the calculation formula is: Where e is the base of the natural logarithm, j is the imaginary unit, and j satisfies 2 = -1, y(n) is the windowed acoustic signal, Y(k) is the frequency domain signal corresponding to y(n), and N is the number of Fourier transform points; Step 1.5: Process the spectrum Y(k) obtained in step 1.4 through a logarithmic Mel filter to obtain the final output logarithmic Mel spectrum feature X. The calculation formula is: X=log(ω M ·||FFT(y)|| 2 ) Among them, y=[y(0),y(1),...,y(N-1)], ω M ∈R d×B is a Mel filter bank, R represents the real number domain, B represents each discrete frequency component of the spectrum obtained by FFT, and d represents the number of Mel filters.

3. The automatic classification method based on the time characteristics of machine sounds as claimed in claim 2, characterized in that: Step 2 is as follows: Step 2.1: After processing in step 1, a logarithmic Mel spectrum feature X is generated, defined as X∈R d×t =[x1,x2,...,x t ], where x i ∈R d×1 ,i∈[1,2,...,t], t represents the frame number; Step 2.2, two frames x in the spectrogram i and x j The similarity between them is expressed as: s ij =similarity(x i ,x j ),i≠j,i,j∈[1,2,...,t] Among them, similarity(x i ,x j ) represents the calculation of the similarity measure between any two feature vectors; Step 2.3, x i The similarity coefficient sim between the entire spectrum i It is expressed as:

4. The automatic classification method based on the time characteristics of machine sounds as claimed in claim 3, characterized in that: Step 3 is as follows: After processing in step 2.3, the similarity coefficient of each frame in the spectrogram is obtained, and then the variance of the similarity coefficient in the entire spectrogram is calculated, which is expressed as: Among them, μ is the mean of the similarity coefficient, σ 2 is the variance of the similarity coefficient.

5. The automatic classification method based on the time characteristics of machine sounds as claimed in claim 4, characterized in that: Step 4 is as follows: All sound signals in the data set of each machine ID are processed through steps 1-3, and the variance of the similarity coefficient of each machine ID data set is calculated. It is expressed as: Where m is the number of sound samples in the dataset for each machine ID.

6. The automatic classification method based on the time characteristics of machine sounds as claimed in claim 5, characterized in that: Step 5 is as follows: After obtaining the similarity coefficient variance set in step 4, each similarity coefficient variance is compared with the empirical variance to determine the stationarity of each sound signal. The specific decision function is expressed as: Where δ is the empirical variance threshold.

7. The automatic classification method based on the time characteristics of machine sounds as claimed in claim 6, characterized in that: Step 6 is as follows: Step 6.1, based on the decision in step 5, obtain the stationarity or non-stationarity determination result of all sound signals in the data set of each machine ID, which is specifically expressed as: Category i ={S,N,S,S,N,....,S,N},i∈[00,01,02,03,04,05,06] Among them, S indicates that it is judged as a stable sound signal, and N indicates that it is judged as a non-stationary sound signal; Step 6.2: For each machine ID, count the number of stationary sound signals and the number of non-stationary sound signals in its data set, denoted as N S and N N ; Step 6.3, based on step 6.2, calculate the proportion of samples determined to be stationary signals and non-stationary signals in the machine ID data set, which are defined as the stationary ratio SR and the non-stationary ratio NR, respectively, specifically expressed as: Among them, SR+NR=1; Then the calculated stability ratio SR is compared with the set threshold θ to determine the overall stability of the machine ID, which is specifically expressed as:

8. An automatic classification system based on the time characteristics of machine sounds, used to execute the method according to any one of claims 1 to 7, characterized in that: Includes the following modules: Logarithmic Mel spectrum feature extraction module: extracts the logarithmic Mel spectrum feature of the original sound signal to obtain the logarithmic Mel spectrum feature X; Similarity coefficient acquisition module: Calculate the similarity of the obtained logarithmic Mel spectrum feature X, by measuring each frame x i With other frames x j The similarity coefficient between each frame and the entire logarithmic Mel spectrum is obtained; where i≠j; Similarity coefficient variance calculation module of the spectrum graph: for the obtained similarity coefficient, the similarity coefficient variance of the spectrum graph is calculated; Similarity coefficient variance calculation module for all sound signals: processes the collected industrial equipment sound signal data set and calculates the similarity coefficient variance of all sound signals in each machine ID data set; Sound signal stability judgment module: compares each similarity coefficient variance obtained with the empirical variance to determine whether each sound signal is stable; Overall stability judgment module: Based on the judgment results of all sound signals in each machine ID data set, the number of signals judged as stable and non-stationary is counted, and the proportion is calculated. The calculated stability ratio is compared with the set threshold θ, and the overall stability of the machine ID is judged based on the comparison results.

Images