Ventilator failure warning system based on data analysis

Abstract

The application discloses a kind of ventilation machine fault early warning systems based on data analysis, it is related to fault early warning technical field, for solving the ventilation machine fault early warning system for mine ventilation machine fault diagnosis there is greater improvement space problem, the present application includes multi-source data perception module, data interface module, abnormal diagnosis module and warning module;Source data perception module is used to collect and perceive the audio signal and surrounding environment data of ventilation machine operation;The present application is based on audio signal analysis, in combination with the running parameters of ventilation machine, the running state of ventilation machine is perceived in all directions, and audio signal adaptive filtering model, feature extraction and abnormal diagnosis model are carried, to ensure that the system is efficiently and stably operated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault early warning, in particular to a ventilation machine fault early warning system based on data analysis. BACKGROUND

[0002] Coal resources are one of the most important basic energy sources and an important factor in ensuring national energy security. With the increase in the number of coal mining by coal enterprises, high-speed cutting coal to increase production brings a large number of coal dust particles floating in the air in the mine roadway, requiring ventilation equipment to timely disperse toxic and harmful gases and coal dust in the mine to ensure the flow of fresh air in the mine.

[0003] Since the ventilation operation needs to be synchronized with the coal mining operation, it is necessary to ensure the long-term stable operation of the ventilation machine, which has a high requirement on the health status of each key equipment in the ventilation machine. After long-time operation of the mine ventilation machine, various types of faults are likely to occur, which seriously threatens the life safety of miners due to the inability to discharge toxic gases in the mine. Therefore, it is necessary for professional technicians to timely find hidden abnormalities in the ventilation machine equipment during routine maintenance.

[0004] However, the mine ventilation machine is always in a dynamic ventilation process, and the occurrence position and time of various faults during operation are uncertain, which increases the difficulty of fault detection.

[0005] The common ventilation machine state fault monitoring system generally completes the sensing of the operating state parameters of the ventilation machine system by setting vibration, pressure, temperature, current and other sensors, but ignores the detection of sound signals generated by each part of the equipment during the operation of the ventilation machine. Once the equipment shows signs of failure, the vibration and temperature signals may not change significantly in a short period of time, while the sound signal is the first to show abnormal fluctuations. Without detection and analysis of the audio signal, it is impossible to realize the comprehensive analysis of the non-contact detection and contact detection data of the ventilation machine, resulting in inaccurate ventilation machine fault analysis results.

[0006] Moreover, there is still a lot of room for improvement in the current mine ventilation machine fault diagnosis technology, mainly in that the fault diagnosis scheme needs to be optimized. The traditional ventilation machine fault diagnosis monitors the operating parameters of different components through various types of sensors and judges whether the ventilation machine has failed by comparing with the threshold value. This way not only has low fault diagnosis accuracy, but also consumes a lot of manpower and lacks intelligence.

[0007] In order to solve the above defects, the present application provides a technical solution. SUMMARY

[0008] The purpose of the present application is to solve the technical problem that the current ventilation fan fault early warning system has a large improvement space for mine ventilation fan fault diagnosis, and a ventilation fan fault early warning system based on data analysis is proposed.

[0009] The purpose of the present application can be realized by the following technical solutions:

[0010] A ventilation fan fault early warning system based on data analysis comprises:

[0011] A multi-source data sensing module is used to collect and sense the ventilation fan operation audio signal and the surrounding environment data;

[0012] A data interface module is used to preliminarily process and preprocess the audio signal and the surrounding environment data; the audio signal is digitally sampled, and an intelligent filtering and noise reduction model is used to preliminarily process the obtained audio signal; the pre-processing method of the audio signal pre-emphasis, frame division and windowing is used for the audio signal after noise reduction and filtering;

[0013] An abnormality diagnosis module is used to segment the obtained ventilation fan audio signal and real-time operation parameter, diagnose the abnormal signal by comparison; the process of diagnosing through the audio signal comprises the following steps:

[0014] S1: the audio signal after noise reduction, filtering, frame division and windowing is divided into several frames, and each frame is a linearly continuous signal;

[0015] S2: then the kurtosis value in the characteristic parameter of the time domain analysis method is used as the real-time monitoring index of the audio signal, different kurtosis threshold values are set, the audio data stream is calculated in real time, the kurtosis value of the audio data stream is calculated, and the kurtosis threshold values are compared, so as to realize the real-time monitoring of the early abnormality of the ventilation fan;

[0016] S3: when the kurtosis abnormal value appears, it is judged whether the abnormality occurs and the type of the abnormality, the audio signal in a period of time before and after the kurtosis abnormal value is saved, the frequency domain processing and time-frequency processing method of the audio signal is used to further process the audio signal in the period of time, and the frequency distribution in the period of time and the time point of the abnormal frequency are obtained;

[0017] The abnormality diagnosis module is also used to apply the CNN network in deep learning to the ventilation fan abnormality diagnosis, and construct a fault diagnosis model;

[0018] Further, the sensing of the ventilation fan operation audio signal by the multi-source data sensing module is composed of multiple directional aluminum tape pickups, the pickups are respectively directed to the ventilation fan impeller, the non-driven side bearing and the driving side motor safe operation part, and the audio signal in the ventilation fan operation process is collected in real time;

[0019] The pickup is annularly distributed around the ventilator, and each pickup is located on a straight line parallel to the bearing of the ventilator and is equidistantly distributed.

[0020] Further, the data interface module adopts the following specific operation steps for the pre-processing of the noise-reduced and filtered audio signal:

[0021] The pre-processing method for the noise-reduced and filtered audio signal is composed of audio signal pre-emphasis and frame division and windowing.

[0022] The audio signal pre-emphasis highlights the high-frequency signal, and the pre-emphasized signal is frame-divided and windowed.

[0023] Further, the specific operation steps of the time-domain analysis method in the abnormal diagnosis module during the diagnosis process of the audio signal are as follows:

[0024] The audio signal after noise reduction, filtering, frame division and windowing is divided into several frames, and the kurtosis value of each frame signal is calculated.

[0025] Further, the specific operation steps of the frequency domain processing method in the abnormal diagnosis module during the diagnosis process of the audio signal are as follows:

[0026] The frequency characteristic information of the signal is obtained by analyzing the amplitude-frequency characteristic and the phase-frequency characteristic of the audio signal.

[0027] B1: Calculate the kurtosis value of each frame of audio signal, and compare it with the pre-set kurtosis threshold value;

[0028] B2: When the kurtosis is greater than the pre-set kurtosis threshold value, record the time point of the kurtosis abnormal moment, and save the audio signal in a period of time before and after the time point of the kurtosis abnormal moment;

[0029] B3: Read the audio signal obtained in step B2, and analyze the audio signal by using the fast Fourier transform method to obtain the frequency distribution of the audio signal in the period of time;

[0030] B4: observe the frequency distribution characteristics of the audio signal, and compare with the audio signal when the ventilator is running normally, if the frequency distribution range obtained twice is the same, the ventilator does not have obvious abnormality; if there is a significant difference in frequency, the ventilator has an abnormality, and the abnormal part of the ventilator is judged according to the size of the frequency energy.

[0031] Further, the specific operation steps of the time-frequency analysis method in the abnormal diagnosis module during the diagnosis process by the audio signal are as follows:

[0032] The time-frequency analysis method is used to process the audio signal to obtain the rule of frequency size changing with time and the time of abnormal frequency appearing, and the short-time Fourier transform method is selected as the time-frequency analysis method;

[0033] The local spectrum concept is introduced in the short-time Fourier transform, a window function with very short time length is used to intercept a segment of the original non-stationary signal, the Fourier transform is performed on the stationary signal, then the window function is slid along the time axis to obtain the image of the frequency of the whole non-stationary signal changing with time, that is, the time-frequency diagram;

[0034] From the frequency components contained in the audio signal and the energy size of each frequency in the obtained time-frequency diagram, the brighter the color of the frequency band in the diagram, the larger the energy of the frequency band, and the larger the proportion of the frequency band in the overall energy. The frequency variation rule corresponding to different time is obtained.

[0035] Further, the specific operation steps of the fault diagnosis implementation process in the abnormal diagnosis module are as follows:

[0036] First, the audio signal acquisition program is transplanted in the microprocessor in the pickup, and the audio signals of different parts of the ventilator during operation are collected in real time; then the data stream formed by the obtained audio signal is processed in real time, the algorithm program of this part is deployed after the audio acquisition program, which is the time-frequency diagram of the audio signal based on the short-time Fourier transform, and then the time-frequency diagram is used to judge whether the ventilator has a fault; at the same time, the neural network is used to train the time-frequency diagram under different working conditions offline, and then the trained network model is used for online automatic judgment of whether the ventilator has an abnormality and the type of the abnormality.

[0037] Further, the specific operation steps of the warning module sending the warning information group to the collection terminal of the manager are as follows:

[0038] The abnormal diagnosis result, occurrence time and specific position of the ventilator are packaged to form a warning information group and sent to the mobile terminal of the manager;

[0039] The mobile phone terminal is provided with a WeChat applet, when receiving the warning program, the mobile phone terminal will vibrate and sound prompt, warning the management personnel to check the mobile phone at the first time, the WeChat applet will display the ventilation fan abnormal diagnosis result, occurrence time and specific position; when the sound or vibration continues for a preset time and has not stopped, information and telephone call are sent to the preset emergency contact person;

[0040] After the warning is completed, the influence time after the ventilation fan abnormal diagnosis warning is recorded, and the ventilation fan abnormal diagnosis result, occurrence time, specific position and response time are recorded according to the number, recorded in the form of a list, and the recorded list file is divided into three screening methods of time screening, abnormal type screening and position screening to view the history file.

[0041] Compared with the prior art, the beneficial effects of the present application are:

[0042] (1) The present application, by constructing a multi-source data perception module, using multi-intelligent sensors to perform multi-source multi-dimensional real-time perception on the running state of the mine ventilation fan, using distributed pickups and environmental data acquisition, distributed around the key operating parts of the ventilation fan, real-time capturing the real-time running state of the ventilation fan, compared with the conventional vibration, temperature and other conventional equipment state detection means, audio detection has its unique advantages, since noise is the product of most mechanical equipment operation, which contains rich information related to the mechanical state during equipment operation;

[0043] (2) The present application, after denoising and preprocessing of the original audio signal, the internal characteristics of the audio signal can be extracted by using signal processing method, the common method for audio signal feature extraction includes time domain feature parameter processing, frequency domain feature parameter and time-frequency analysis, while the present system can simultaneously apply the three methods to audio signal processing according to actual field demand and real-time of the system, and the advantages and disadvantages of different methods are integrated, the features of the collected audio signal are extracted, so that more detailed and accurate audio feature information is obtained;

[0044] (3) The present application, by using the set warning module, the warning information package can be sent to the mobile phone terminal of the management personnel at the first time when the ventilation fan fails, and the attention of the management personnel to the mobile phone is improved through the sound and vibration of the mobile phone, so that the ventilation fan failure can be understood at the first time, and the next step of maintenance or repair can be carried out. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to facilitate the understanding of those skilled in the art, the present application will be further described below with reference to the drawings;

[0046] Figure 1 The system block diagram of the present application;

[0047] Figure 2 This is a flowchart of the kurtosis calculation process for audio signals in this invention;

[0048] Figure 3 This is a diagram illustrating the fault diagnosis process of the fault diagnosis model in this invention;

[0049] Figure 4 This is a structural diagram of the CNN network in this invention. Detailed Implementation

[0050] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] It should be understood that the terms “comprising” and “including” used in this disclosure and claims indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0052] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure. As used in this disclosure and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this disclosure and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0053] like Figures 1-4 As shown, a ventilator fault early warning system based on data analysis includes a multi-source data sensing module, a data interface module, an anomaly diagnosis module, and an alert module.

[0054] The multi-source data sensing module is used to collect and sense the audio signals of the ventilator operation and the surrounding environmental data;

[0055] The system for sensing the audio signals of the ventilator operation consists of multiple directional aluminum ribbon microphones. The microphones are pointed at the ventilator impeller, the non-drive side bearing, and the key parts of the drive side motor for safe operation, respectively, to collect the audio signals during the operation of the ventilator in real time. The microphones are arranged in a ring around the ventilator, and all microphones are located on a straight line parallel to the ventilator bearing. Each microphone is equidistant from the others, and each microphone is pointed at the key operating parts of the ventilator, thus achieving all-round perception of the ventilator's audio signals.

[0056] The sensing of the peripheral environment data is composed of temperature sensors and smoke sensors, which sense the temperature and smoke parameters around the ventilator;

[0057] The data receiving module is used for the preliminary processing and preprocessing of the real-time ventilator operation audio signals and peripheral environment data uploaded by the multi-source data sensing module;

[0058] The audio signals are digitally sampled according to the required signal sampling rate, and the collected digital audio signals are preliminarily processed by using an intelligent filtering and noise reduction model to improve the signal-to-noise ratio of the audio signals;

[0059] The PCM1808 audio processing chip is specifically used to complete the accurate sampling of the audio signals, and the process is to convert the continuous analog signals into discrete digital signals through three steps of signal sampling, quantization and coding:

[0060] Signal sampling: a number of representative sample values are taken out from a continuously changing analog signal in time to represent the continuously changing analog signal, and a function of an analog audio signal which is continuous in time and amplitude is denoted as x(t), and the sampling process is to discretize the function x(t) in time;

[0061] Quantization is to approximately represent the originally continuously changing amplitude value in time with a limited number of amplitudes, so as to change the continuous amplitude of the analog signal into discrete values with a certain time interval;

[0062] Coding is to represent the discrete values after quantization with binary codes according to a certain rule. In order to ensure the high fidelity of the audio signals on site, the PCM coding format is usually used;

[0063] The intelligent filtering and noise reduction model is used to improve the signal-to-noise ratio of the audio signals, the accuracy of feature extraction of the signals and the operation processing speed, and to realize adaptive, self-learning, efficient and stable filtering process for background noise in different environments;

[0064] The pre-processing method for the audio signals after noise reduction filtering is composed of audio signal pre-emphasis and frame division and windowing; the audio signal pre-emphasis can highlight the high-frequency signals and reduce the interference of the low-frequency signals on the high-frequency signals, and the formula of the pre-emphasis digital filter before digital conversion is:

[0065] H(z) = 1 - αz -1

[0066] The value is between 0 and 1 and closer to 1 is better, and the value is 0.98, and the amplitude-frequency characteristic is:

[0067]

[0068] When the signal is high frequency, the amplitude |H(w)| tends to 1+α, and low frequency tends to 1-α, and the high frequency part can be highlighted after pre-emphasis processing;

[0069] The pre-emphasized signal is framed and windowed, and after the audio signal is framed, the frequency domain information is missing. In order to reduce the influence of spectrum loss, windowing is also used. Windowing is to multiply the original sound signal by a window function, and then analyze and study the windowed signal. When the sound recognition uses the Hamming window (Hamming Window) time domain function, its function form can be expressed as:

[0070]

[0071] Where k=1, 2, …, N, and the frequency domain characteristic expression of the Hamming window is:

[0072]

[0073] The receiving of the environment data around the ventilator adopts a wired and wireless heterogeneous communication network for receiving;

[0074] The abnormal diagnosis module divides the obtained audio signal and the real-time running parameter of the ventilator, calculates the running characteristics of the ventilator, and compares the abnormal signals for diagnosis;

[0075] The process of diagnosing through the audio signal includes the following steps:

[0076] S1: The audio signal after noise reduction filtering and framing and windowing is divided into several frames, and each frame is a linear continuous change signal;

[0077] S2: Then the kurtosis in the characteristic parameter of the time domain analysis method is used as the real-time monitoring index of the audio signal, different kurtosis thresholds are set, the audio data stream is calculated in real time, the kurtosis value of the audio data stream is calculated, and compared with different kurtosis thresholds, the real-time monitoring of the early abnormality of the ventilator is realized;

[0078] S3: When the kurtosis abnormal value appears, it is judged whether the abnormality occurs and the type of the abnormality, the audio signal in a period of time before and after the kurtosis value abnormal moment is saved, the frequency distribution in the period of time and the time point of the abnormal frequency are obtained by further processing the audio signal in the period of time using the frequency domain processing and time-frequency processing method of the audio signal;

[0079] The time domain analysis method in step S2 is the most accurate, fast and direct method in signal processing, which analyzes the abnormal conditions of the electromechanical equipment by calculating the time domain statistical characteristics of the audio signal, judges the running state of the electromechanical equipment, and can also divide the health status grade of the ventilator according to the historical monitoring data and maintenance records. Common time domain indicators include skewness, kurtosis, variance, maximum value, minimum value, peak value, square root amplitude, mean amplitude, root mean square amplitude, absolute mean value, waveform index, amplitude index, pulse index, margin index and kurtosis index. According to the characteristics of the audio signal and the meaning represented by different time domain parameters, the kurtosis parameter in the time domain processing method is selected as the time domain parameter processing; the kurtosis reflects the distribution characteristics of the signal, and the kurtosis coefficient Ku increases with the increase of the proportion of amplitude abnormal pulse in the signal. As a mathematical statistics variable, the kurtosis value has a certain regular distribution. When the ventilator is in normal operation, the overall change of the audio signal is relatively uniform, and there is no abnormal impact signal. The probability density distribution of different amplitude signals can be approximated as a normal distribution, and the kurtosis coefficient Ku of the normal distribution oscillates around the value 3, and there is no large abnormal value. When the motor or the rolling bearing has surface scratches or damage, the audio signal generated during operation will have a large impact, that is, the amplitude abnormal sound signal. The probability density of the amplitude abnormal pulse in the signal increases with the aggravation of the fault, and the probability density distribution of different amplitude signals gradually deviates from the normal distribution. The more serious the fault, the larger the kurtosis coefficient of the audio signal. When it exceeds the threshold value, it needs to be alarmed for maintenance, which can be used as an important basis for early fault diagnosis of the ventilator. The expression of kurtosis calculation is: In the formula, x(t) is a signal with zero mean, and p(x) is the probability density distribution function of x(t); the kurtosis index value is independent of the speed, size and load of the ventilator driving motor or rolling bearing, and is mainly used for early abnormal form detection of rolling bearings and driving motors;

[0080] Specifically, the audio signal after noise reduction filtering and frame windowing is divided into several frames, and then the kurtosis value of each frame signal is calculated. The audio acquisition frequency is set to 16000, and the audio data stream in 1s is divided into 30 frames according to the frame principle. The number of kurtosis values calculated in one second is 30, which meets the demand of real-time monitoring of the ventilator. The kurtosis calculation process of the audio signal can be referred to as shown in Figure 2

[0081] ​The frequency domain processing method in S3 is to obtain the frequency characteristic information of the signal by analyzing the amplitude-frequency characteristic and the phase-frequency characteristic of the audio signal; the frequency domain analysis completes the Fourier transform of the audio signal, analyzes the frequency characteristic of the signal, and adopts the fast FFT algorithm. The Fourier transform is a commonly used means for analyzing the frequency domain characteristics of a signal. In the field of digital signal processing, the fast Fourier transform (FFT) is improved on the basis of the discrete Fourier transform (DFT), reduces the time complexity of the operation, and greatly speeds up the development of digital signal processing. The specific process of adopting the fast Fourier transform (FFT) as the frequency domain processing method of the audio signal is as follows:

[0082] B1: Calculate the kurtosis value of each frame of audio signal, and compare it with the pre-set kurtosis threshold;

[0083] B2: If the kurtosis is greater than the pre-set kurtosis threshold, record the time point of the kurtosis value abnormal moment, and save the audio signal in a period of time before and after the time point to the specified path of the local;

[0084] B3: Read the audio signal, and analyze the audio signal by using the FFT method to obtain the frequency distribution of the audio signal in the period of time;

[0085] B4: Observe the frequency distribution characteristics of the audio signal, and compare it with the audio signal when the ventilator is normally running. If the frequency distribution ranges obtained at two times are the same, the ventilator does not have obvious abnormalities. On the contrary, if there is a significant difference in frequency, the ventilator has an abnormality, and the abnormal part of the ventilator is judged according to the size of the frequency energy.

[0086] The time-frequency processing method in step S3 is to process the audio signal by using the time-frequency analysis method to obtain the law of frequency size changing with time and the time of abnormal frequency occurrence, and to select the short-time Fourier transform method as the time-frequency analysis method. In the short-time Fourier transform, the concept of local spectrum is introduced, a window function with very short time length is used to intercept a segment of the original non-stationary signal, the Fourier transform of the stationary signal is performed, and then the window function is slid along the time axis to obtain the image of the frequency of the entire non-stationary signal changing with time. For a continuous signal s(t), the definition of its continuous short-time Fourier transform (STFT) is:

[0087]

[0088] Its inverse transform is:

[0089]

[0090] In the formula, g(t) is a window function with very short time length; * represents complex conjugate. When g(t) = 1, the short-time Fourier transform is converted into the traditional Fourier transform.

[0091] From the obtained time-frequency diagram, not only can the frequency components contained in the audio signal and the energy size of each frequency be clearly seen, the brighter the color band in the diagram, the greater the energy of the band, the greater the proportion of the band in the overall energy, and at the same time, the frequency variation law corresponding to different time can be obtained, which is convenient for finding the time of abnormal frequency of the ventilator, and has important significance for abnormal diagnosis of the ventilator;

[0092] The feature extraction for abnormal diagnosis of the ventilator uses the common method in signal processing, and combines the advantages and disadvantages of different methods. The features of the collected audio signal are extracted, so that more detailed and accurate audio feature information is obtained;

[0093] The abnormal diagnosis module is also used to apply the CNN network in deep learning to the abnormal diagnosis of the ventilator, and construct a fault diagnosis model. The fault diagnosis process of the model is shown in Figure 3 ;

[0094] The fault diagnosis implementation process of the fault diagnosis model is as follows:

[0095] First, the audio signal acquisition program is transplanted in the microprocessor in the pickup, which is used to collect the audio signals of different components of the ventilator in real time. Then the collected audio data stream is processed in real time. This part of the algorithm program is deployed after the audio acquisition program, which is mainly the audio signal time-frequency diagram based on short-time Fourier transform. Since the time-frequency diagram under different working conditions of the ventilator has certain differences, it can be judged whether the ventilator has failed according to the time-frequency diagram. Then the neural network is used to train the time-frequency diagram under different working conditions offline, and then the trained network model is used to automatically judge whether the ventilator has an abnormality and the type of the abnormality. The key to realizing this part is the feature extraction of the audio signal and the selection and parameter setting of the neural network;

[0096] The type of neural network used and its parameter setting, the commonly used convolutional neural network (CNN) is selected, which is used to automatically identify different types of time-frequency diagrams, forming a short-time Fourier transform and CNN combined ventilator abnormal automatic diagnosis model. The working process of the model is as follows:

[0097] First, the kurtosis is used for multi-source signal processing. The audio signal in the adjacent time of the kurtosis value at the abnormal time is further analyzed by using FFT and short-time Fourier transform. Since the result obtained by FFT transform is not conducive to differentiation, the time-frequency diagram obtained by short-time Fourier transform is selected as the input of the CNN model. Through convolution, pooling and full connection process, deeper feature information is obtained. Finally, the classification function is used to classify the output result to obtain the abnormal category of the ventilator;

[0098] The accuracy of the fault diagnosis model is related to the size of the input data and the parameter design of the CNN network. For the parameter design of the CNN network:

[0099] Input layer: The size of the input data affects the operation time and fault diagnosis accuracy of the model. The frequency distribution in the short-time Fourier spectrum of the audio signal of the ventilator under different working conditions is mainly concentrated in the low frequency region. The input data is segmented and compressed to obtain a 32x32 pixel image.

[0100] Convolutional layer: The convolutional layer is the core part of the CNN network. This layer obtains the local receptive field of the input data by a set of convolution kernels with the same size. In feature extraction, a convolution kernel can only extract part of the data features. Therefore, multiple convolution kernels are needed to extract all the features of the input data. The dimensions of the convolution kernel correspond to the dimensions of the input data. Since the Hilbert spectrum is two-dimensional data, a two-dimensional convolution kernel is used to extract features by convolution calculation. For the selection of the convolution kernel, the noise of the original audio signal has been reduced before feature extraction, so the noise effect of convolution calculation can be ignored. Therefore, the size of the first convolution kernel is selected as 5x5. Since the pooling layer can cause some feature information to be lost, the second convolution layer uses more convolution kernels with a size of 3x3.

[0101] Pooling layer: The pooling layer is used to collect part of the features of the convolutional layer and reduce the data dimension. Its working principle is to divide the data of the convolutional layer into different regions according to the preset size, and perform maximum pooling and average pooling according to the selected pooling type. Maximum pooling selects the maximum value in the region as the feature value, while average pooling calculates the average value of the region as the feature information. Since the pooling layer does not have convolution operation, it will not increase the depth of the feature map. Through the pooling operation, the dimension of the feature map is greatly reduced, the calculation speed of the network model is accelerated, and the network overfitting phenomenon is avoided. For the design of the CNN pooling layer, the maximum pooling method is adopted, the size is 2x2, the step is 2, and the data padding method is selected as same.

[0102] Activation function: The activation function is a kind of non-linear mapping function, which is used with the convolutional layer to enhance the expression ability of the model. The original multilayer perceptron model cannot learn data features well without the introduction of the activation function. The introduction of the activation function in the CNN network increases the nonlinearity of the model and accelerates the gradient descent speed of the network training, which is used to process a large number of nonlinear data problems.

[0103] Fully connected layer: The data processed by the convolutional layer and the pooling layer is finally transmitted to the fully connected layer. In order to avoid overfitting, the Dropout regularization mechanism is introduced in the fully connected layer to effectively suppress a part of neurons during network training and only update the remaining neurons.

[0104] Working mechanism: the CNN network training adopts the back propagation mode, the principle is that the output result generated by the forward training is compared with the actual value, if the output deviation is large, the convolution kernel size, learning rate, data set ratio and other parameters need to be adjusted, so as to obtain the most suitable parameter value, this working mode based on output result adjusting network hyperparameters is called back propagation, using this working mode in the designed fault diagnosis model can improve the model output accuracy; the designed CNN structure refers to Figure 4 ;

[0105] The warning module is used for receiving the ventilation fan abnormal diagnosis result output by the abnormal diagnosis module, and acquiring the occurrence time and specific position of the ventilation fan abnormal diagnosis result;

[0106] The ventilation fan abnormal diagnosis result, occurrence time and specific position are sent to the mobile terminal of the management personnel, and a warning program is packaged and sent;

[0107] The mobile terminal is provided with a WeChat applet, when receiving the warning program, the mobile terminal will vibrate and sound, prompting the management personnel to check the mobile phone at the first time, and the WeChat applet will display the ventilation fan abnormal diagnosis result, occurrence time and specific position;

[0108] When the sound or vibration lasts for a preset time and does not stop, information and a call are sent to the preset emergency contact person, so as to avoid that the management personnel do not check the mobile phone;

[0109] After completing the warning, the influence time after the ventilation fan abnormal diagnosis warning is recorded, and the ventilation fan abnormal diagnosis result, occurrence time, specific position and response time are recorded according to the number, and the recorded list file can be quickly viewed in history by time filtering, abnormal type filtering and position filtering.

[0110] The preferred embodiments disclosed above are only used to help explain the present application. The preferred embodiments do not describe all the details, nor limit the application to the specific implementation. Obviously, according to the content of the specification, many modifications and changes can be made. The specification selects and describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application. The application is limited by the claims and their entire scope and equivalents.

Claims

1. A data analysis based ventilation fan failure warning system, characterized in that, The application relates to a ventilation machine early abnormality real-time monitoring system, which comprises the following: A multi-source data sensing module for collecting and sensing ventilation machine running audio signals and peripheral environment data; A data interface module for preliminarily processing and preprocessing the audio signals and the peripheral environment data; the audio signals are digitally sampled, and an intelligent filtering and noise reduction model is used to preliminarily process the obtained audio signals; A preprocessing method for the audio signals after noise reduction filtering, which comprises audio signal pre-emphasis, framing and windowing; An abnormality diagnosis module for segmenting the obtained ventilation machine audio signals, comparing and diagnosing abnormal signals; the diagnosis process of the audio signals comprises the following steps: S1: the audio signals after noise reduction filtering, framing and windowing are divided into a plurality of frames, and each frame is a linearly continuous signal; S2: then, the kurtosis value in the characteristic parameters of the time domain analysis method is used as a real-time monitoring index of the audio signals, different kurtosis threshold values are set, audio data streams are obtained by real-time calculation of the audio signals, the kurtosis values of the audio data streams are calculated, and the kurtosis values are compared with the different kurtosis threshold values, so that the ventilation machine early abnormality is monitored in real time; S3: when the kurtosis abnormal value appears, whether the abnormality occurs and the type of the abnormality are judged, the audio signals in a period of time before and after the kurtosis value abnormal moment are saved, the audio signals in the period of time are further processed by using a frequency domain processing method and a time-frequency analysis method, and the frequency distribution in the period of time and the time point of the abnormal frequency are obtained; The abnormality diagnosis module is also used for applying a CNN network in deep learning to ventilation machine abnormality diagnosis to construct a fault diagnosis model; An alarm module for receiving the ventilation machine abnormality diagnosis result output by the abnormality diagnosis module and sending an alarm information group to a management personnel's collection terminal; The specific operation steps of the frequency domain processing method in the diagnosis process of the audio signals are as follows: The frequency characteristic information of the signal is obtained by analyzing the amplitude-frequency characteristic and the phase-frequency characteristic of the audio signal; the fast Fourier transform is used as the frequency domain processing method of the audio signal after the frequency domain analysis is completed, and the specific process is as follows: B1: the kurtosis value of each frame of audio signal is calculated and compared with the pre-set kurtosis threshold value; B2: when the kurtosis is greater than the pre-set kurtosis threshold value, the time point of the kurtosis value abnormal moment is recorded, and the audio signals in a period of time before and after the time point of the kurtosis value abnormal moment are saved; B3: the audio signals obtained in step B2 are read, and the fast Fourier transform method is used to analyze the audio signals, so that the frequency distribution of the audio signals in the period of time is obtained; B4: the frequency distribution characteristics of the audio signals are observed, and the audio signals during normal operation of the ventilation machine are used as comparison, if the frequency distribution ranges obtained twice are the same, the ventilation machine does not have obvious abnormality; if there is obvious difference in the frequency, the ventilation machine has abnormality, and the abnormal position of the ventilation machine is judged according to the frequency energy size.

2. A data analysis based ventilation machine failure warning system as claimed in claim 1, wherein, The sensing of the ventilation machine running audio signals by the multi-source data sensing module is composed of a plurality of directional aluminum tape pickups; the pickups are respectively directed to the ventilation machine fan impeller, the non-driven side bearing and the driving side motor safe operation position, and the audio signals in the ventilation machine running process are collected in real time. The pickup devices are annularly distributed around the ventilator, and each pickup device is located on a straight line parallel to the bearing of the ventilator and is equidistantly distributed.

3. A data analysis based ventilation machine failure warning system as claimed in claim 1, wherein, The specific operation steps of the pre-processing method for the noise-reduced audio signal are as follows: The pre-processing method for the noise-reduced audio signal is composed of audio signal pre-emphasis and frame division and windowing; The audio signal pre-emphasis highlights high-frequency signals, and the pre-emphasized signal is frame-divided and windowed.

4. A data analysis based ventilation machine failure warning system as claimed in claim 1, wherein, The specific operation steps of the time-domain analysis method in the diagnosis process of the audio signal by the abnormal diagnosis module are as follows: The audio signal after noise reduction filtering and frame division and windowing is divided into several frames, and then the kurtosis value of each frame signal is calculated.

5. A data analysis based ventilation machine failure warning system as claimed in claim 1, wherein, The specific operation steps of the time-frequency analysis method in the diagnosis process of the audio signal by the abnormal diagnosis module are as follows: The time-frequency analysis method is used to process the audio signal to obtain the law of frequency change with time and the time of abnormal frequency occurrence, and the short-time Fourier transform method is selected as the time-frequency analysis method. The short-time Fourier transform introduces the concept of local spectrum, uses a window function with very short time length to intercept a segment of the original non-stationary signal, performs Fourier transform on the segment, then slides the window function along the time axis to obtain the image of the frequency change rule of the entire non-stationary signal with time, i.e. the time-frequency diagram. From the obtained time-frequency diagram, the frequency components contained in the audio signal and the energy size of each frequency are obtained, and the brighter the color band in the diagram, the greater the energy of the color band, and the greater the proportion of the color band in the overall energy.

6. A data analysis based ventilation machine failure warning system as claimed in claim 1, wherein, The specific operation steps of the fault diagnosis implementation process in the abnormal diagnosis module are as follows: First, the audio signal acquisition program is transplanted in the microprocessor in the pickup device to collect the audio signals of different components of the ventilator in real time; then the obtained audio signal is formed into a data stream and processed in real time, the fault diagnosis model is deployed after the audio acquisition program, which is based on the time-frequency diagram of the audio signal by short-time Fourier transform to determine whether the ventilator has failed; at the same time, the neural network is used to train the time-frequency diagram under different working conditions, and then the trained fault diagnosis model is used for online automatic judgment of whether the ventilator has failed and the type of failure.

7. A data analytics based ventilation machine failure warning system as claimed in claim 1, wherein, The specific operation steps of the warning information group sent by the warning module to the collection terminal of the manager are as follows: The abnormal diagnosis result, occurrence time and specific position of the ventilator are packaged into a warning information group and sent to the mobile terminal of the manager; When receiving the warning program, the mobile phone terminal vibrates and gives a sound prompt, and the abnormal diagnosis result of the ventilator, the occurrence time and the specific location are displayed; when the sound or vibration continues for a preset time and has not stopped, information and a call are sent to the preset emergency contact person; After completing the warning, the influence time after the ventilator abnormal diagnosis warning is recorded, and the ventilator abnormal diagnosis result, the occurrence time, the specific location and the response time are recorded according to the number, recorded in the form of a list, and the recorded list file is divided into three filtering methods of time filtering, abnormal type filtering and location filtering for viewing of historical files.

Images