A neural network-based device fault sound source positioning method

Abstract

The application discloses a kind of based on neural network equipment fault sound source positioning method, the method includes using microphone to carry out acquisition to sound source information, establishes probability function calculation and the same type equipment fault sound source appears probability;Using neural network to carry out feature learning to new equipment fault sound source;Optimize convolution layer in neural network, improve training speed;Establish sound source acquisition function and utilize two microphones simultaneously to carry out acquisition to fault sound source;Using fusion function to the correlation operation of fault sound source collected by two microphones;Using position coordinate function to solve to fault sound source, obtains the specific position of fault sound source, the method is using existing neural network model to be applied in equipment fault sound source positioning field, greatly shorten the time consumed by artificial search fault point, the positioning result of the method is accurate, operability is strong, help to improve the production efficiency of enterprise.

Description

Technical Field

[0001] This invention relates to the field of equipment fault diagnosis and localization, and in particular to a method for locating the sound source of equipment faults based on neural networks. Background Technology

[0002] Equipment is prone to malfunction during prolonged operation, especially rotating equipment. When these devices malfunction, they typically produce abnormal noise. Currently, faulty equipment is detected through manual inspection. However, manual inspection often fails to identify faulty equipment immediately, leading to delays in addressing problems and potentially disrupting normal production.

[0003] Patent publication number CN204945337U discloses a fault location and alarm protection device for electromechanical equipment. This device includes: a starting module for starting a motor and / or mechanical equipment; a fault location and alarm protection module; and a temperature monitoring module. The fault location and alarm protection module is used to locate the fault upon receiving a temperature alarm signal from the temperature monitoring module and to control the starting module to cut off power to protect the motor and / or mechanical equipment. The temperature monitoring module monitors the casing temperature of the motor and / or mechanical equipment and sends a temperature alarm signal to the fault location and alarm protection module when the casing temperature exceeds a set threshold. Using this device, maintenance personnel can be remotely notified of the fault status, and the operation of the electromechanical equipment can be automatically stopped, preventing overheating and burnout.

[0004] Patent CN113884809A discloses a method and system for secondary equipment fault location based on neural networks, comprising: acquiring fault feature information of secondary equipment, wherein the fault feature information includes operating status information, SV / GOOSE / MMS receiving status information, and voltage and current sampling values; determining whether the fault feature information of the secondary equipment can be used to infer fault location through knowledge base information; if so, performing fault location based on the fault feature information of the secondary equipment and the knowledge base information; if not, inputting the fault feature information of the secondary equipment into a trained fault location model to obtain the power grid fault location result, wherein the trained fault location model is obtained by training an LSTM-RNN network with historical fault information of the secondary equipment.

[0005] However, some equipment is currently very large, and disassembling it is time-consuming and laborious, making it impossible to accurately locate the fault using existing methods. But when the equipment malfunctions, it will produce abnormal sounds, and the source of the sound can be used to locate the fault. Summary of the Invention

[0006] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides a method for locating the sound source of device faults based on neural networks.

[0007] The technical solution adopted in this invention is that the method includes the following steps:

[0008] Step S1: Collect sound source information using a microphone, and establish a probability function to calculate the probability of the occurrence of the same type of equipment fault sound source;

[0009] Step S2: Use a neural network to learn the features of new equipment fault sound sources;

[0010] Step S3: Optimize the convolutional layers in the neural network to improve training speed;

[0011] Step S4: Establish a sound source acquisition function to simultaneously acquire fault sound sources using two microphones;

[0012] Step S5: Use the fusion function to perform correlation operations on the fault sound sources collected by the two microphones;

[0013] Step S6: Solve the fault sound source using the position coordinate function to obtain the specific location of the fault sound source.

[0014] Furthermore, the probability function is established to calculate the probability of the occurrence of fault sound sources of the same type of equipment. The calculation formula is as follows:

[0015]

[0016] Among them, A n q represents the probability of a device malfunction sound source occurring. n p represents a single device fault sound source in the feature set. m This represents the set of all device fault sound sources in the feature set.

[0017] Furthermore, the feature learning for the new equipment fault sound source is calculated using the following formula:

[0018]

[0019] Among them, c n Indicates a new source of equipment malfunction, r n q represents the output of feature learning for new equipment fault sound sources using a neural network. n p represents a single device fault sound source in the feature set. m This represents the set of all device fault sound sources in the feature set.

[0020] Furthermore, the convolutional layer is optimized, and the calculation formula is as follows:

[0021]

[0022] Where d(i) represents the judgment output of the convolutional layer in the neural network, i represents the number of convolutional layers, e represents the natural constant, N represents the number of pooling layers, and σ represents the convolution coefficient.

[0023] Furthermore, the calculation formula for collecting the fault sound source is as follows:

[0024]

[0025] Among them, f x (z) represents the sound signal collected by microphone 1, h(z) represents the sound signal emitted by the device during operation, and j x (z) represents the noise signal collected by microphone 1, g x This indicates the sound collection coefficient of microphone #1;

[0026]

[0027] Among them, f y (z) represents the sound signal collected by microphone 2, where z is the variance of the sound signal per unit time, μ. xy j represents the error in the sound signal collected by microphone 1 and microphone 2. y (z) represents the noise signal collected by microphone #2, g y This represents the sound collection coefficient of microphone #2.

[0028] Furthermore, the fusion function is calculated using the following formula:

[0029]

[0030] Among them, T xy (μ xy ) represents the fusion result, and K represents the fusion coefficient.

[0031] Furthermore, the fault sound source is solved using a position coordinate function, and the calculation formula is as follows:

[0032]

[0033] Where, r l The x-coordinate representing the microphone position, s l The vertical coordinate t represents the microphone position. lThe vertical coordinates of the microphone position are represented by w, r1 represents the horizontal coordinate of the faulty sound source 1 position, s1 represents the vertical coordinate of the faulty sound source 1 position, t1 represents the vertical coordinate of the faulty sound source 1 position, u1 represents the distance between the faulty sound source 1 position and the microphone position, r2 represents the horizontal coordinate of the faulty sound source 2 position, s2 represents the vertical coordinate of the faulty sound source 2 position, t2 represents the vertical coordinate of the faulty sound source 2 position, u2 represents the distance between the faulty sound source 2 position and the microphone position, r3 represents the horizontal coordinate of the faulty sound source 3 position, s3 represents the vertical coordinate of the faulty sound source 2 position, t3 represents the vertical coordinate of the faulty sound source 3 position, u3 represents the distance between the faulty sound source 3 position and the microphone position.

[0034] Beneficial effects:

[0035] This invention proposes a method for locating the sound source of equipment faults based on neural networks. It utilizes two microphones to collect sound information, employs a neural network to learn sound features, determines whether the sound source is abnormal, fuses the sound sources collected by the two microphones using a fusion function, and finally calculates the sound source location using a position coordinate function. This method applies existing neural network models to the field of equipment fault sound source localization, significantly reducing the time spent manually searching for fault points. The method provides accurate localization results, is highly operable, and helps improve enterprise production efficiency. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the overall steps of the present invention;

[0037] Figure 2 This is a schematic diagram illustrating the location of the fault sound source according to the present invention. Detailed Implementation

[0038] It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. The following describes the application in further detail with reference to the accompanying drawings and specific embodiments.

[0039] like Figure 1 As shown, a method for locating the sound source of a device fault based on a neural network includes the following steps:

[0040] Step S1: Collect sound source information using a microphone, and establish a probability function to calculate the probability of the occurrence of the same type of equipment fault sound source;

[0041] The microphone array uses a microphone array to collect sound sources related to equipment malfunctions. The microphone array consists of a number of acoustic sensors (usually microphones) that sample and process the spatial characteristics of the sound field. Its main functions include sound source localization, suppression of background noise, interference, reverberation, and echo, and signal extraction and separation. Sound source localization refers to calculating the angle between the sound source and the array using the microphone array, and tracking the target sound source based on TDOA (Time Difference of Arrival). Signal extraction and separation refers to effectively forming a beam in the desired direction and picking up only the signal within the beam, thereby simultaneously extracting the sound source and suppressing noise. Furthermore, using information provided by the microphone array based on deep neural networks, effective reverberation removal can be achieved, greatly improving the effect of voice interaction in real-world application scenarios.

[0042] This microphone array employs a planar distribution structure, containing three microphones, enabling 360-degree equivalent sound pickup with a wake-up resolution of 1 degree. The microphone array can be used to acquire raw and noise-reduced frequencies, determine the wake-up angle, and identify the main microphone.

[0043] Reference signal interface: can be used for echo cancellation;

[0044] Microphone numbers: corresponding to 0-5 respectively;

[0045] LED numbering: Starting with microphone number 0, the LEDs are numbered clockwise from 0 to 5;

[0046] The audio type acquired by the microphone and the wake-up angle are defined as follows:

[0047] Noise reduction frequency: sampling rate 16kHz, 16bit, one channel;

[0048] Original audio: sampling rate 16kHz, 32-bit eight channels, of which channels 1-3 correspond to 3 microphones, and channels 4-5 are reference signals;

[0049] Wake-up angle: Starting from microphone 0, clockwise the angles are 0-35°.

[0050] The microphone array board's J1 connector connects to the host computer's USB interface via a USB communication cable. It uses a standard USB Type-A interface, operating at 5V and 0.5A. This is used for audio data transmission and control signal transmission. The microphone array board's J2 connector connects to the host computer's power amplifier via a 2-pin reference signal cable, specifically to the speaker's SPK+ and SPK- pins. This is used for transmitting the echo cancellation reference signal, and the peak-to-peak value of the input reference signal must not exceed 300mV.

[0051] A probability function describes the probability of a device malfunction sound source occurring. It involves the sound source of a single device malfunction in the feature set as well as the set of all device malfunction sound sources. Through the probability function, the frequency of different device malfunctions can be determined. The sound source of a single device malfunction is the average of the features collected from the sound source multiple times, while the set of all device malfunction sound sources refers to the average of the features of all device malfunction sound sources.

[0052] Step S2: Use a neural network to learn the features of new equipment fault sound sources;

[0053] The neural network used in this invention is a convolutional neural network, which is used to learn the features of frequency, noise, and peak-to-peak values ​​of all sound sources.

[0054] Step S3: Optimize the convolutional layers in the neural network to improve training speed;

[0055] The function of a convolutional layer is to extract features from input data. It contains multiple convolutional kernels, each element of which corresponds to a weight coefficient and a bias, similar to a neuron in a feedforward neural network. Each neuron in a convolutional layer is connected to multiple neurons in a nearby region in the previous layer. The size of this region depends on the size of the convolutional kernel and is often called the "receptive field," analogous to the receptive fields of visual cortex cells. During operation, the convolutional kernel systematically scans the input features, performing element-wise matrix multiplication and summing within the receptive field, and then adding the bias. Convolutional layer parameters include kernel size, stride, and padding, which together determine the magnitude of the output features and are hyperparameters of the convolutional neural network. The kernel size can be specified as any value smaller than the input data size; a larger kernel allows for more complex extracted input features. In this invention, the convolutional layer is optimized to reduce the computation time for feature learning, thereby improving training speed.

[0056] like Figure 2 As shown, step S4: Establish a sound source acquisition function to simultaneously acquire fault sound sources using two microphones;

[0057] In this invention, the two microphones refer to two microphone arrays. When a faulty sound source appears, the two microphone arrays will collect different sound source characteristics because they are at different distances from the faulty sound source.

[0058] Step S5: Use the fusion function to perform correlation operations on the fault sound sources collected by the two microphones;

[0059] The fusion function is related to the acquisition results of the two microphones. This step provides reliable sound source information for step S6 to prevent the microphones from acquiring different faulty sound sources, thus avoiding the overlap of faulty sound sources.

[0060] Step S6: Solve the fault sound source using the position coordinate function to obtain the specific location of the fault sound source.

[0061] The essence of the position coordinate function in this invention is to solve for the difference in coordinates, and to finally locate the position of the fault sound source by using the coordinate difference between the fault sound source and the microphone position.

[0062] A probability function is established to calculate the probability of the occurrence of fault sound sources in similar types of equipment. The calculation formula is as follows:

[0063]

[0064] Among them, A n q represents the probability of a device malfunction sound source occurring. n p represents a single device fault sound source in the feature set. m This represents the set of all device fault sound sources in the feature set.

[0065] Feature learning is performed on the new equipment fault sound source, and the calculation formula is as follows:

[0066]

[0067] Among them, c n Indicates a new source of equipment malfunction, r n q represents the output of feature learning for new equipment fault sound sources using a neural network. n p represents a single device fault sound source in the feature set. m This represents the set of all device fault sound sources in the feature set.

[0068] The convolutional layer is optimized using the following formula:

[0069]

[0070] Where d(i) represents the judgment output result of the convolutional layer in the neural network, i represents the number of convolutional layers, e represents the natural constant, N represents the number of pooling layers, and σ represents the convolution coefficient.

[0071] The sound source of the fault is collected, and the calculation formula is as follows:

[0072]

[0073] Among them, f x (z) represents the sound signal collected by microphone 1, h(z) represents the sound signal emitted by the device during operation, and j x (z) represents the noise signal collected by microphone 1, g x This indicates the sound collection coefficient of microphone #1;

[0074]

[0075] Among them, f y (z) represents the sound signal collected by microphone 2, where z is the variance of the sound signal per unit time, μ. xy j represents the error in the sound signal collected by microphone 1 and microphone 2. y (z) represents the noise signal collected by microphone #2, g y This represents the sound collection coefficient of microphone #2.

[0076] The fusion function is calculated using the following formula:

[0077]

[0078] Among them, T xy (μ xy ) represents the fusion result, and K represents the fusion coefficient.

[0079] The fault sound source is solved using a position coordinate function, and the calculation formula is as follows:

[0080]

[0081] Where, r l The x-coordinate representing the microphone position, s l The vertical coordinate t represents the microphone position. l The vertical coordinates of the microphone position are represented by w, r1 represents the horizontal coordinate of the faulty sound source 1 position, s1 represents the vertical coordinate of the faulty sound source 1 position, t1 represents the vertical coordinate of the faulty sound source 1 position, u1 represents the distance between the faulty sound source 1 position and the microphone position, r2 represents the horizontal coordinate of the faulty sound source 2 position, s2 represents the vertical coordinate of the faulty sound source 2 position, t2 represents the vertical coordinate of the faulty sound source 2 position, u2 represents the distance between the faulty sound source 2 position and the microphone position, r3 represents the horizontal coordinate of the faulty sound source 3 position, s3 represents the vertical coordinate of the faulty sound source 2 position, t3 represents the vertical coordinate of the faulty sound source 3 position, u3 represents the distance between the faulty sound source 3 position and the microphone position.

[0082] This invention proposes a method for locating the sound source of equipment faults based on neural networks. It utilizes two microphones to collect sound information, employs a neural network to learn sound features, determines whether the sound source is abnormal, fuses the sound sources collected by the two microphones using a fusion function, and finally calculates the sound source location using a position coordinate function. This method applies existing neural network models to the field of equipment fault sound source localization, significantly reducing the time spent manually searching for fault points. The method provides accurate localization results, is highly operable, and helps improve enterprise production efficiency.

[0083] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for locating the sound source of a device fault based on a neural network, characterized in that, The method includes the following steps: Step S1: Collect sound source information using a microphone, and establish a probability function to calculate the probability of the occurrence of the same type of equipment fault sound source; Step S2: Use a neural network to learn the features of new equipment fault sound sources; Step S3: Optimize the convolutional layers in the neural network to improve training speed; Step S4: Establish a sound source acquisition function to simultaneously acquire fault sound sources using two microphones; Step S5: Use the fusion function to perform correlation operations on the fault sound sources collected by the two microphones; Step S6: Solve the position coordinate function for the fault sound source to obtain the specific location of the fault sound source; The probability function is established to calculate the probability of the occurrence of fault sound sources of the same type of equipment. The calculation formula is as follows: ， Among them, A n q represents the probability of a device malfunction sound source occurring. n p represents a single device fault sound source in the feature set. m This represents the set of all device fault sound sources in the feature set; The feature learning for the new equipment fault sound source is calculated using the following formula: ， Among them, c n Indicates a new source of equipment malfunction, r n q represents the output of feature learning for new equipment fault sound sources using a neural network. n p represents a single device fault sound source in the feature set. m This represents the set of all device fault sound sources in the feature set; The convolutional layer is optimized, and the calculation formula is as follows: ， Where d(i) represents the judgment output of the convolutional layer in the neural network, i represents the number of convolutional layers, e represents the natural constant, N represents the number of pooling layers, and σ represents the convolution coefficient; The calculation formula for collecting the fault sound source is as follows: ， in, h(z) represents the sound signal collected by microphone 1, h(z) represents the sound signal emitted by the device during operation, and j x (z) represents the noise signal collected by microphone #1. This indicates the sound collection coefficient of microphone #1; ， in, z represents the variance of the sound signal collected by microphone #2 per unit time. j represents the error in the sound signal collected by microphone 1 and microphone 2. y (z) represents the noise signal collected by microphone #2. This indicates the sound collection coefficient of microphone #2; The fusion function is calculated using the following formula: ， in, This represents the fusion result, where K represents the fusion coefficient. The fault sound source is solved using a position coordinate function, and the calculation formula is as follows: ， Where, r l The x-coordinate representing the microphone position, s l The vertical coordinate t represents the microphone position. l The vertical coordinates of the microphone position are represented by w, r1 represents the horizontal coordinate of the faulty sound source 1 position, s1 represents the vertical coordinate of the faulty sound source 1 position, t1 represents the vertical coordinate of the faulty sound source 1 position, u1 represents the distance between the faulty sound source 1 position and the microphone position, r2 represents the horizontal coordinate of the faulty sound source 2 position, s2 represents the vertical coordinate of the faulty sound source 2 position, t2 represents the vertical coordinate of the faulty sound source 2 position, u2 represents the distance between the faulty sound source 2 position and the microphone position, r3 represents the horizontal coordinate of the faulty sound source 3 position, s3 represents the vertical coordinate of the faulty sound source 2 position, t3 represents the vertical coordinate of the faulty sound source 3 position, u3 represents the distance between the faulty sound source 3 position and the microphone position.

Images