Multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system and method

Abstract

The invention belongs to the technical field of industrial equipment fault detection application, and particularly discloses a multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system and method, and the system comprises a multi-channel audio acquisition module, an audio preprocessing and spectrogram conversion module, a weighted fusion module, a fault judgment and decision module, and an alarm push module. The identification method comprises the following steps: S1, multi-channel audio acquisition; s2, audio preprocessing and spectrogram conversion; s3, carrying out weighted fusion; s4, fault judgment and decision making; and S5, alarm pushing. The beneficial effects of the invention are that: 1) the multi-channel audio acquisition device comprehensively acquires sound signals of key parts of the coal mill, avoids information omission, can capture unique sound features of faults of different parts, provides rich data for accurate judgment, and improves the recognition capability; and 2) the weight fusion module based on the attention mechanism dynamically adjusts the weight according to the fault importance, highlights the key part information, and improves the fault detection accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial equipment fault detection application, and specifically relates to a multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system and method, which specifically belongs to the multidisciplinary cross-fields of mechanical engineering, signal processing, artificial intelligence, etc., and is mainly used for status monitoring and fault diagnosis of large rotating mechanical equipment such as coal mills in thermal power generation, mining and other industries. By collecting, analyzing and processing the sound signals of coal mill operation, accurate identification and early warning of coal mill faults are achieved, so as to improve the reliability, safety and operation efficiency of the equipment and reduce maintenance costs and downtime. Background Art

[0002] In industrial production, coal mill is one of the important equipment, and its stable operation is crucial to the smooth progress of the production process. However, coal mill is prone to failure due to various factors such as wear, fatigue, overload, etc. during long-term operation. Traditional coal mill fault detection methods mainly include regular maintenance, vibration monitoring, etc., but these methods have certain limitations.

[0003] Regular maintenance requires downtime, which affects production efficiency and may not detect faults in time. Although vibration monitoring can detect equipment anomalies to a certain extent, it may not be sensitive to some early faults or specific types of faults. In addition, traditional methods often require the installation of a large number of sensors, which is costly and complex to install and maintain.

[0004] With the continuous development of science and technology, voiceprint recognition technology has gradually been applied to the field of industrial equipment fault detection. Sound signals contain rich information about the equipment's operating status. By collecting and analyzing the sound of coal mill operation, real-time monitoring and diagnosis of equipment failures can be achieved. However, the existing coal mill voiceprint recognition methods still need to be improved in terms of accuracy, reliability and practicality.

[0005] At present, on the one hand, the existing voiceprint recognition methods usually adopt a single audio collection method, which makes it difficult to fully obtain the sound information of different parts of the coal mill, resulting in limited coverage of fault detection; on the other hand, in the process of sound signal processing, there is a lack of effective denoising algorithms and intelligent fusion technology, which makes the noise interference larger and affects the accuracy of fault identification. At the same time, when using sound signals for fault diagnosis, the existing methods often use simple pattern recognition algorithms, which are difficult to capture the complex patterns and time dynamic changes of sound signals, resulting in low accuracy of fault identification.

[0006] In summary, in order to overcome the shortcomings of existing coal mill fault detection methods and improve the accuracy, reliability and practicality of fault detection, research and development solutions are urgently needed.

[0007] Based on the above problems, the present invention provides a multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system and method. Summary of the invention

[0008] Purpose of the invention: The purpose of the present invention is to provide a multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system and method to solve many problems existing in the existing coal mill fault detection technology, improve the accuracy of fault detection, and provide an efficient and reliable fault monitoring solution for the field of coal mill status monitoring.

[0009] Technical solution: The first aspect of the present invention provides a multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system, including a multi-channel audio acquisition module, an audio preprocessing and spectrogram conversion module, a weighted fusion module, a fault judgment and decision module and an alarm push module, the multi-channel audio acquisition module, the audio preprocessing and spectrogram conversion module, the weighted fusion module, the fault judgment and decision module and the alarm push module are connected in sequence; the multi-channel audio acquisition module is used to comprehensively capture the sound signals of the coal mill during operation, provide original sound data for subsequent processing, and collect sound data using multi-channel audio acquisition devices arranged at key positions of the coal mill main bearing, motor, auger gear and reducer; the audio preprocessing and spectrogram conversion module is used to process the sound signals collected by the multi-channel audio acquisition module, and use traditional audio denoising algorithm to process the collected audio to remove background noise. The interference of sound is eliminated, and then the processed audio is converted into a spectrogram, and the characteristics of the spectrogram are analyzed to provide a high-quality data basis for subsequent analysis and fusion; the weighted fusion module is used to extract the features of the spectrograms of each part after being processed by the audio preprocessing and spectrogram conversion modules, and to perform weighted fusion according to the importance of the fault, so as to integrate the spectrogram information of different parts into a fused spectrogram that can more comprehensively reflect the operating status of the coal mill; the fault judgment and decision module is used to process the fused spectrogram after being processed by the weighted fusion module, use a graph structure to represent the frequency and time relationship in the spectrogram, and use a graph neural network method to analyze and process the fused spectrogram to determine whether the coal mill has a fault and the location and type of the fault; the alarm push module is used to push the information processed by the fault judgment and decision module, and push the fault information to the designated platform and the corresponding operation and maintenance personnel.

[0010] A second aspect of the present invention provides a multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method, comprising the following steps:

[0011] S1. Multi-channel audio acquisition: Multi-channel audio acquisition devices are arranged at key parts of the coal mill such as the main bearing, motor, auger gear and reducer to fully capture the sound signals of the coal mill during operation;

[0012] S2, audio preprocessing and spectrogram conversion: use traditional audio denoising algorithm to process the collected audio, convert the processed audio into a spectrogram, and analyze the characteristics of the spectrogram;

[0013] S3, weighted fusion: through the weighted fusion module, the features of the spectrograms of each part are extracted and weighted fusion is performed according to the importance of the fault;

[0014] S4. Fault judgment and decision-making: Use a graph structure to represent the frequency and time relationship in the spectrogram, and use methods such as graph neural networks to analyze and process the fused spectrogram to determine whether there is a fault in the coal mill and the location and type of the fault;

[0015] S5. Alarm push: Push the coal mill fault to the corresponding platform to remind the operation and maintenance personnel to repair it in time.

[0016] As a preferred embodiment, the arrangement position and number of the multi-channel audio acquisition device are determined through precise calculation and testing to ensure that the sound signals at all angles can be fully captured. For example, for the main bearing of the coal mill, at least three microphones are installed at different positions of the bearing seat, close to the axial and radial directions respectively, and the microphone arrangement satisfies formula (1):

[0017] N main_bearing =α×S bearing (1)

[0018] Among them, N main_bearing Indicates the number of microphones at the main bearing (unit: piece), S bearing Main bearing surface area (single

[0019] : square meters), α is a coefficient determined by experiments, and its value range is [0.1, 0.3].

[0020] As a preferred embodiment, traditional audio denoising algorithms include Wiener filtering algorithm, spectral subtraction and adaptive comb filtering.

[0021] Wave

[0022] When the Wiener filter algorithm is used, the formula 2) is satisfied:

[0023]

[0024] Among them, y(n) is the audio signal after denoising, is the estimated pure audio signal, For pure audio signal

[0025] Difference, is the variance of the noise signal.

[0026] As a preferred implementation, the weighted fusion module dynamically adjusts the weights of the spectrograms of each part according to the importance of the fault based on the attention mechanism to fuse the spectrogram features of each part. The weight calculation formula (3) is:

[0027] W i =f(I i ,θ)(3),

[0028] Among them, W i represents the weight of the spectrogram of the ith part, I i is the importance index of the part, which can be obtained through the statistics of historical fault data, and θ is the parameter set of the attention mechanism;

[0029] The fusion formula (4) is: F = ∑ i W i S i (4), where F is the fused spectrogram, S i is the spectrogram of the ith part.

[0030] As a preferred implementation, the use of a graph structure to represent the frequency and time relationship in the spectrogram is to regard each frequency point in the spectrogram as a node in the graph.

[0031] The connections between nodes are determined based on the correlation between frequencies and the order in time, as follows:

[0032] a. For frequency correlation, let two frequency points f i and f j , if their correlation coefficient r ij Greater than the preset correlation threshold r th , then a connection is established in the graph, namely A ij =1 (where A is the adjacency matrix, A ij represents the element in the row and column of the matrix), otherwise A ij = 0. Correlation coefficient r ij It can be obtained by calculating the Pearson correlation coefficient or other suitable correlation measurement methods between the spectrogram feature vectors corresponding to two frequency points.

[0033] b. For the time sequence, within the same time frame, node connections are constructed in the order of frequency; for different time frames, if the time interval is less than the preset time interval threshold t th , and the corresponding frequency points have a certain correlation (the correlation coefficient satisfies the above r th Condition), a connection is also established.

[0034] c. The feature vector X of each node iIt is obtained by extracting the features of the corresponding frequency points in the spectrogram. For example, information such as frequency amplitude and phase can be extracted as elements of the feature vector. These feature vectors will be used as the initial state information of the nodes in the graph neural network for subsequent information propagation and processing.

[0035] As a preferred implementation, the recursive neural network model based on the gated graph neural network includes an input layer, a hidden layer, and an output layer, which are closely related to the constructed graph structure, as follows:

[0036] a. The input layer receives the fused spectrogram data and converts it into a format suitable for graph neural network processing, that is, the spectrogram data is mapped according to the relationship between graph nodes and edges defined in the constructed graph structure to obtain the node feature vector set V input For example, for each frequency point corresponding to a node, its amplitude, phase and other characteristic values ​​in the spectrogram are combined into a vector as the initial characteristic vector of the node.

[0037] b. The hidden layer uses a gating mechanism to update the node status. Formula (5) is:

[0038] H t =f(A,V t-1 ,X)(5),

[0039] Among them, H t is the hidden state at time t (i.e., the updated node feature vector set), A is the adjacency matrix (if a graph structure is used to represent the frequency and time relationship in the spectrogram, each frequency point in the spectrogram is regarded as a node in the graph, and the connection between nodes is determined according to the correlation between frequencies and the order in time, which describes the connection relationship between nodes), V t-1 is the node state at the previous moment, X is the input feature vector (related to the node feature vector in the input layer, including the currently input spectrogram information), and f is the gating function. The gating function can adopt the gating mechanism in the common gated recurrent unit (GRU) or long short-term memory network (LSTM), which effectively captures the spatiotemporal dynamic changes of the sound signal in the graph structure by controlling the flow and forgetting of information.

[0040] c. The output layer is based on the final state H of the hidden layer. final Output the coal mill fault judgment result. Formula (6) can be expressed as:

[0041] Y=g(H final ) (6),

[0042] Among them, g is the output function, for example, it can be a fully connected layer plus a softmax function, which is used to classify and predict the fault category of the coal mill, and output whether the fault exists and the probability distribution of the fault location and type.

[0043] As a preferred embodiment, the multi-stage decision includes determining whether there is a possibility of a fault based on the probability value output by the network. If the probability value exceeds a certain threshold, it is considered that a fault may exist, and further analyzing the characteristics of the spectrogram to determine the location where the fault may occur, and combining historical fault data and expert knowledge to accurately judge the type of fault. Suppose the fault probability output by the network is P, when P ≥ T (where T is a preset threshold), further analysis is initiated. The analysis of the spectrogram features can use feature extraction algorithms, such as principal component analysis (PCA), etc., to extract key features for determining the fault location. The way to combine historical fault data can be to build a fault database and determine the possible fault type by similarity calculation and other methods.

[0044] As a preferred implementation, when extracting features from a spectrogram, a Mel-frequency cepstral coefficient (MFCC) method may be used, the steps of which include framing and windowing the audio signal, then performing a fast Fourier transform to obtain a spectrum, then performing a Mel filter group to obtain a Mel spectrum, and finally calculating the logarithmic energy and performing a discrete cosine transform to obtain an MFCC feature vector.

[0045] The third aspect of the present invention provides an electronic device and a computer-readable storage medium, wherein the electronic device includes a memory, a processor, and a multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition program stored in the memory and runnable on the processor, and when the processor executes the multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method program, the steps of the multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method are implemented.

[0046] The computer-readable storage medium stores a program of a multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system identification method. When the multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system identification method is executed by a processor, the steps of the multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system identification method are implemented.

[0047] Compared with the prior art, the beneficial effects of the multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system and method of the present invention are: 1) The multi-channel audio acquisition device comprehensively obtains the sound signals of the key parts of the coal mill to avoid information omission, and can capture the unique sound characteristics of faults in different parts, providing rich data for accurate judgment, and improving the ability to identify complex faults; 2) The weighted fusion module based on the attention mechanism dynamically adjusts the weight according to the importance of the fault, highlights the information of key parts, and integrates to form a fused spectrogram that more accurately reflects the overall operating status, thereby improving the accuracy of fault detection; 3) The graph structure combines the gated neural network to analyze the frequency and time relationship of the spectrogram, which can better capture complex spatiotemporal patterns and dynamic changes, and timely discover early weak fault signals. The present invention provides an innovative solution for the field of coal mill status monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] 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 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 these drawings without paying creative work.

[0049] Figure 1 This is a system structure diagram of the present invention

[0050] Figure 2 It is a flow chart of the system identification method of the present invention. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0052] In the description of the present invention, it should be noted that the orientations or positional relationships indicated by the terms "top", "bottom", "one side", "the other side", "front", "back", "middle part", "inside", "top end", "bottom end", etc. are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention; the terms "first", "second", and "third" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance; in addition, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it 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 it can be an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0053] Embodiment 1

[0054] like Figure 2 The implementation method of the identification system of the present invention shown in the figure specifically comprises the following steps:

[0055] Step 1: Multi-channel audio acquisition:

[0056] a. Choose a microphone with high sensitivity and strong anti-interference ability as the audio acquisition device. For example, you can choose a professional industrial-grade microphone, whose frequency response range should be able to cover the sound frequency range generated by the operation of the coal mill (generally 20Hz-20kHz or even wider).

[0057] b. For the main bearing of the coal mill, microphones are evenly arranged around the bearing seat, such as one installed every 90 degrees, one near the axial and radial positions, to ensure that the sound signals of the main bearing during operation can be fully captured. On the surface of the motor housing, microphones are installed at several key locations, such as the front cover, rear cover, and side of the motor, to obtain the sound of the motor running. For auger gears and reducers, microphones are reasonably arranged near key transmission parts according to their structural characteristics.

[0058] c. The microphone is fixed to the corresponding part of the coal mill through a special bracket or clamp to ensure a firm installation to avoid displacement or loosening of the microphone due to factors such as coal mill vibration. At the same time, attention should be paid to protecting the microphone to prevent dust, oil and other factors from affecting its performance. Dust covers and protective covers can be used.

[0059] d. Set a suitable sampling frequency, generally 44.1kHz or 48kHz can be selected to ensure that the details of the coal mill sound signal can be accurately collected. At the same time, according to the operating characteristics of the coal mill and the needs of fault detection, determine the duration and interval of the collection. For example, you can collect continuously for a period of time (such as 1 hour), and then collect once every certain period of time (such as 5 minutes) to monitor the operating status of the coal mill in real time and detect faults in time.

[0060] e. Calibrate the microphone of each channel to ensure the consistency of its sensitivity and frequency response. You can use a standard audio signal generator to generate a sound signal of known frequency and amplitude, test and calibrate each microphone, and adjust its gain and other parameters to make the collected signal accurate and reliable.

[0061] Step 2: Audio preprocessing and spectrogram conversion:

[0062] a. When using spectral subtraction for denoising, firstly, the collected audio signal is framed, and the frame length is generally selected to be 20-40 milliseconds. Then, each frame signal is subjected to a fast Fourier transform (FFT) to obtain its spectrum. By analyzing the audio signal over a period of time (such as a few seconds), the spectrum of the background noise is estimated. The spectrum of the noisy audio signal is subtracted from the noise spectrum to obtain the spectrum after preliminary denoising, and then an inverse fast Fourier transform (IFFT) is performed to obtain the denoised audio signal.

[0063] b. For Wiener filtering denoising, the power spectral density of the pure audio signal and the noise signal is estimated based on the prior knowledge of the audio signal and the noise or through the preliminary analysis of the collected signal. Then the coefficients of the filter are calculated according to the formula of Wiener filtering, and the audio signal is filtered to obtain the denoised signal.

[0064] c. Use short-time Fourier transform (STFT) to convert the denoised audio into a spectrogram. Divide the audio signal into short-time frames (such as the frame length setting above), perform windowing on each frame (common window functions include Hanning window, Hamming window, etc.), and then perform FFT on the windowed frame to obtain the spectrum of each frame. Arrange the spectra of adjacent frames in chronological order to form a spectrogram, where the horizontal axis represents time, the vertical axis represents frequency, and the energy intensity of the spectrum is represented by color or grayscale.

[0065] d. Use the Mel Frequency Cepstrum Coefficient (MFCC) method to convert the spectrogram. After framing and windowing the audio signal, perform FFT to obtain the spectrum, and then filter the spectrum through a Mel filter bank to convert the frequency scale to the Mel scale. The Mel filter bank usually consists of multiple bandpass filters, whose center frequencies are distributed according to the Mel scale. Calculate the logarithmic energy of the filtered signal, and then perform discrete cosine transform (DCT) to obtain the MFCC feature vector, which can be used as a feature representation of the spectrogram.

[0066] Step 3: Weighted fusion:

[0067] a. According to the coal mill historical failure data and expert experience, determine the relative importance of each part when a failure occurs. For example, if the main bearing failure has a greater impact on the operation of the coal mill and has a higher frequency of occurrence in historical failures, then give the main bearing a higher importance weight. The importance of each part can be quantitatively evaluated using methods such as the analytic hierarchy process (AHP) to obtain the importance index I of each part. i .

[0068] b. Importance index I i Substitute into the weight calculation formula W i =f(I i ,θ) to calculate the weight. Here f can be a simple linear function or a nonlinear function, and θ is the parameter of the function, which can be adjusted and optimized through training data. For example, (linear weighting), or use a neural network model to transform I i As input, the weight W is obtained after the neural network is trained i .

[0069] c. Extract the features of the spectrogram of each part, and use methods such as principal component analysis (PCA) to extract the main features. Then, according to the calculated weight W i , weighted fusion of the features of the spectrograms of each part. For example, suppose there are two parts whose spectrogram feature vectors are S1 = [s 11 ,s 12 ,…,s 1n ] and S2 = [s 21 ,s 22 ,…,s 2n ], whose corresponding weights are W1 and W2 respectively, then the fused feature vector F = [W1s 11 +W2s 21 ,W1s 12 +W2s 22 ,…,W1s 1n +W2s 2n ].

[0070] Step 4: Fault diagnosis and decision making:

[0071] a. For frequency correlation, calculate the correlation coefficient (such as Pearson correlation coefficient) between the feature vectors corresponding to each frequency point in the spectrogram. If the correlation coefficient of two frequency points is greater than a preset threshold (such as 0.6), they are considered to be correlated in frequency and connected in the graph.

[0072] b. For the time sequence, within the same time frame, node connections are built in order of frequency. For different time frames, if the time interval is less than a preset threshold (such as 50 milliseconds) and the corresponding frequency points are related, connections are also established.

[0073] c. The feature vector of each node may include information such as frequency amplitude, phase, energy, etc. For example, the amplitude value, phase value of the frequency point and the energy integral value within a certain time window are combined into a multi-dimensional feature vector as the feature vector of the node.

[0074] d. The input layer preprocesses the fused spectrogram data to make it conform to the input format of the graph neural network. For example, the spectrogram data is arranged in the order of graph nodes and the feature vector of each node is used as input.

[0075] e. The hidden layer uses a gated mechanism such as a gated recurrent unit (GRU) or a long short-term memory network (LSTM) to update and propagate node information. By adjusting the parameters of the gated unit, it can effectively learn the spatiotemporal characteristics of the sound signal. For example, the update gate and reset gate in the GRU can selectively update the hidden state of the node based on the current input and the hidden state of the previous moment, capturing the dynamic changes of the sound signal.

[0076] f. The output layer uses a fully connected layer plus a softmax function to convert the output of the hidden layer into a probability distribution of whether a fault exists and the location and type of the fault. For example, the output result may be a vector [p 正常 ,p 主轴承故障 ,p 电机故障 ,…], where p 正常 It represents the probability of normal operation of the coal mill, and the other elements represent the probabilities of different fault types.

[0077] g. Set a fault probability threshold, such as 0.5. When the fault probability output by the network is greater than or equal to the threshold, it is considered that a fault may exist. Then further analyze the spectrogram features, such as calculating the mean, standard deviation, peak value and other statistics of the features, and determine the possible location of the fault based on these statistics.

[0078] h. Combine historical fault data to establish a fault case library. Compare the currently detected spectrogram features with the features in the historical fault case library, and use similarity measurement methods (such as Euclidean distance, cosine similarity, etc.) to find the most similar historical fault cases, thereby determining the fault type. At the same time, combine expert knowledge to verify and correct the judgment results to improve the accuracy of fault diagnosis.

[0079] Step 5: Alarm push:

[0080] a. You can choose the host computer software of the industrial monitoring system as the main push platform to display fault alarm information in real time on the monitoring interface. At the same time, combined with the mobile phone SMS platform and email system, the alarm information will be sent to relevant operation and maintenance personnel.

[0081] b. The alarm information should include the equipment number of the coal mill, the time of the fault, the initial fault location and type, and the recommended treatment measures. For example, "Coal mill #001 detected a possible main bearing failure at [specific time]. Please check and repair it in time."

[0082] c. The system is connected to the mobile phone SMS platform and email system through a programming interface. When a fault is detected, the corresponding interface function is automatically called to send the alarm information to the pre-set recipient. For the host computer software of the industrial monitoring system, the fault alarm information is displayed in the corresponding interface position through the real-time data update mechanism, and the operation and maintenance personnel are reminded by sound, flashing, etc.

[0083] Embodiment 2

[0084] like Figure 1 The identification system of the present invention shown:

[0085] a. Multi-channel audio acquisition module: Multi-channel audio acquisition devices are arranged at key parts of the coal mill such as the main bearing, motor, auger gear and reducer to fully capture the sound signals of the coal mill during operation and provide original sound data for subsequent processing. This module is used to implement the function of step 1 in Example 1 and will not be repeated here;

[0086] b. Audio preprocessing and spectrogram conversion module: Use traditional audio denoising algorithm to process the collected audio.

[0087] Remove interference such as background noise, then convert the processed audio into a spectrogram, and analyze the characteristics of the spectrogram to provide a high-quality data basis for subsequent analysis and fusion. This module is used to implement the function of step 1 in Example 1, and will not be repeated here;

[0088] c. Weighted fusion module: extract features from the spectrograms of each part and perform weighted fusion according to the importance of the fault.

[0089] The spectrogram information of different parts is integrated into a fused spectrogram that can more comprehensively reflect the operating status of the coal mill. This module is used to implement the function of step 3 in embodiment 1, which will not be described in detail here;

[0090] d. Fault judgment and decision module: Use graph structure to represent the frequency and time relationship in the spectrogram, use graph neural network and other methods to analyze and process the fused spectrogram, and determine whether there is a fault in the coal mill and the location and type of the fault. This module is used to implement the function of step 4 in Example 1, and will not be repeated here;

[0091] e. Alarm push module: pushes fault information to the designated platform and corresponding operation and maintenance personnel. This module is used to implement the function of step 5 in embodiment 1, and will not be described in detail here.

[0092] Implementation Case 3

[0093] An electronic device and a computer-readable storage medium of the present invention:

[0094] The electronic device includes a memory, a processor, and a multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition program stored in the memory and runnable on the processor. When the processor executes the multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method program, the steps of the multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method are implemented.

[0095] The computer-readable storage medium stores a program of a multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system identification method. When the multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system identification method is executed by a processor, the steps of the multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system identification method are implemented.

[0096] It should be noted that, in this article, terms such as "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0097] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. Multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system, characterized by: It includes a multi-channel audio acquisition module, an audio preprocessing and spectrogram conversion module, a weighted fusion module, a fault judgment and decision module and an alarm push module, and the multi-channel audio acquisition module, the audio preprocessing and spectrogram conversion module, the weighted fusion module, the fault judgment and decision module and the alarm push module are connected in sequence; The multi-channel audio acquisition module is used to fully capture the sound signals of the coal mill during operation, provide original sound data for subsequent processing, and collect sound signals using multi-channel audio acquisition devices arranged at key parts of the coal mill main bearing, motor, auger gear and reducer; The audio preprocessing and spectrogram conversion module is used to process the sound signal collected by the multi-channel audio acquisition module, use the traditional audio denoising algorithm to process the collected audio, remove the interference of background noise, and then convert the processed audio into a spectrogram, and analyze the characteristics of the spectrogram to provide a high-quality data basis for subsequent analysis and fusion; The weighted fusion module is used to extract features from the spectrograms of each part after being processed by the audio preprocessing and spectrogram conversion modules, and to perform weighted fusion according to the importance of the fault, so as to integrate the spectrogram information of different parts into a fused spectrogram that can more comprehensively reflect the operating status of the coal mill; The fault judgment and decision module is used to process the fused spectrogram after being processed by the weighted fusion module, use a graph structure to represent the frequency and time relationship in the spectrogram, and use a graph neural network method to analyze and process the fused spectrogram to determine whether there is a fault in the coal mill and the location and type of the fault; The alarm push module is used to push the information processed by the fault judgment and decision module, and push the fault information to the designated platform and corresponding operation and maintenance personnel.

2. According to claim 1, the multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method is characterized by: The following steps are included: S1. Multi-channel audio acquisition: Multi-channel audio acquisition devices are arranged at the key parts of the coal mill main bearing, motor, auger gear and reducer to fully capture the sound signals when the coal mill is running; S2, audio preprocessing and spectrogram conversion, using traditional audio denoising algorithm to process the collected audio, and convert the processed audio into a spectrogram, and analyze the characteristics of the spectrogram; S3, weighted fusion, through the weighted fusion module, the characteristics of the acoustic spectrogram of each part are extracted, and weighted fusion is performed according to the importance of the fault; S4, fault judgment and decision-making, using a graph structure to represent the frequency and time relationship in the spectrogram, and using the graph neural network method to analyze and process the fused spectrogram to determine whether the coal mill has a fault and the location and type of the fault; S5. Alarm push: push the coal mill failure to the corresponding platform to remind the operation and maintenance personnel to repair it in time.

3. The multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method according to claim 2 is characterized by: The arrangement position and quantity of the multi-channel audio acquisition device in S1 are determined through precise calculation and testing to ensure that sound signals from all angles can be fully captured; For the main bearing of the coal mill, at least three microphones are installed at different positions of the bearing seat, close to the axial and radial directions respectively. The microphone arrangement must satisfy formula (1): N main_bearing =α×S bearing (1), Among them, N main_bearing Indicates the number of microphones at the main bearing, S bearing is the main bearing surface area, surface area unit is square meters, α is a coefficient determined by experiments, and its value range is [0.1, 0.3].

4. The multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method according to claim 2 is characterized by: The conventional audio denoising algorithm in S2 includes one or more of a Wiener filter algorithm, a spectral subtraction method and an adaptive comb filter; When the Wiener filtering algorithm is used, formula (2) is satisfied: Among them, y(n) is the audio signal after denoising, is the estimated pure audio signal, For pure audio signal Difference, is the variance of the noise signal.

5. The multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method according to claim 2 is characterized by: The weighted fusion module in S3 is based on the attention mechanism and dynamically adjusts the weight of the spectrogram of each part according to the importance of the fault to fuse the spectrogram features of each part. The weight calculation formula (3) is: W i =f(I i ,θ)(3), Among them, W i represents the weight of the spectrogram of the ith part, I i is the importance index of the part, which can be obtained through the statistics of historical fault data, and θ is the parameter set of the attention mechanism; The fusion formula (4) is: F = Σ i W i S i (4), where F is the fused spectrogram, S i is the spectrogram of the ith part.

6. The multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method according to claim 2 is characterized by: The use of a graph structure in S4 to represent the frequency and time relationship in the spectrogram is to regard each frequency point in the spectrogram as a node in the graph; The connections between nodes are determined based on the correlation between frequencies and the order in time, as follows: a. For frequency correlation, let two frequency points f i and f j , if their correlation coefficient r ij Greater than the preset correlation threshold r th , then a connection is established in the graph, namely A ij =1, where A is the adjacency matrix, A ij represents the element in the row and column of the matrix, otherwise A ij =0, correlation coefficient r ij It can be obtained by calculating the Pearson correlation coefficient or other suitable correlation measurement method between the spectrogram feature vectors corresponding to two frequency points; b. For the time sequence, within the same time frame, node connections are constructed in the order of frequency; for different time frames, if the time interval is less than the preset time interval threshold t th , and the corresponding frequency points have a certain correlation, and the correlation coefficient satisfies the above r th If the condition is met, a connection is also established; c. The feature vector X of each node i By extracting the features of the corresponding frequency points in the spectrogram, the frequency amplitude and phase information can be extracted as elements of the feature vector. These feature vectors will be used as the initial state information of the nodes in the graph neural network for subsequent information propagation and processing.

7. The multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method according to claim 6 is characterized by: The recursive neural network model based on the gated graph neural network in S4 includes an input layer, a hidden layer, and an output layer, which are closely related to the constructed graph structure, as follows: a. The input layer receives the fused spectrogram data and converts it into a format suitable for graph neural network processing, that is, the spectrogram data is mapped according to the relationship between graph nodes and edges defined by the constructed graph structure to obtain the node feature vector set V input ; b. The hidden layer uses a gating mechanism to update the node status. Formula (5) is: H t =f(A,V t-1 ,X) (5), Among them, H t is the hidden state at time t, that is, the updated node feature vector set, A is the adjacency matrix, and the constructed graph structure describes the connection relationship between nodes. t-1 is the node state at the previous moment, X is the input feature vector, which is related to the node feature vector in the input layer and contains the spectrogram information of the current input, and f is the gating function. The gating function can adopt the gating mechanism in the common gated recurrent unit (GRU) or long short-term memory network (LSTM) to effectively capture the spatiotemporal dynamic changes of the sound signal in the graph structure by controlling the flow and forgetting of information; c. The output layer is based on the final state H of the hidden layer. final Output the coal mill fault judgment result, which can be expressed by formula (6): Y=g(H final ) (6), Among them, g is the output function, which can be a fully connected layer plus a softmax function, which is used to classify and predict the fault category of the coal mill, and output whether the fault exists and the probability distribution of the fault location and type.

8. The multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method according to claim 2 is characterized by: The S5 alarm push adopts a multi-stage decision-making process, including determining whether there is a possibility of a fault based on the probability value output by the network. If the probability value exceeds a certain threshold, it is considered that a fault may exist. The characteristics of the acoustic spectrogram are further analyzed to determine the possible location of the fault, and the fault type is accurately judged by combining historical fault data and expert knowledge. Assume the failure probability of the network output is P. When P ≥ T, further analysis is initiated, where T is the preset threshold; The feature extraction algorithm is used to analyze the characteristics of the acoustic spectrogram, and the key features are extracted to determine the fault location. The method of combining historical fault data can be to build a fault database and determine the possible fault type through similarity calculation and other methods; The feature extraction algorithm includes but is not limited to the principal component analysis (PCA) algorithm.

9. The multi-channel audio acquisition and intelligent fusion coal mill fault voiceprint recognition system recognition method according to claim 8 is characterized by: When extracting features from the spectrogram, the Mel-frequency cepstral coefficient (MFCC) method can be used. The steps include framing and windowing the audio signal, then performing a fast Fourier transform to obtain the spectrum, then obtaining the Mel spectrum through a Mel filter bank, and finally calculating the logarithmic energy and performing a discrete cosine transform to obtain the MFCC feature vector.

Images