CN112052551B - Fan surge operation fault identification method and system - Google Patents

Fan surge operation fault identification method and system Download PDF

Info

Publication number
CN112052551B
CN112052551B CN201911021826.5A CN201911021826A CN112052551B CN 112052551 B CN112052551 B CN 112052551B CN 201911021826 A CN201911021826 A CN 201911021826A CN 112052551 B CN112052551 B CN 112052551B
Authority
CN
China
Prior art keywords
frequency data
cnn
time
fault
layer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201911021826.5A
Other languages
Chinese (zh)
Other versions
CN112052551A (en
Inventor
翟永杰
杨旭
彭雅妮
王新颖
张磊
华志刚
章义发
李璟涛
吴水木
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Spic Power Operation Technology Institute
State Power Investment Corp ltd
North China Electric Power University
Original Assignee
Spic Power Operation Technology Institute
State Power Investment Corp ltd
North China Electric Power University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Spic Power Operation Technology Institute, State Power Investment Corp ltd, North China Electric Power University filed Critical Spic Power Operation Technology Institute
Priority to CN201911021826.5A priority Critical patent/CN112052551B/en
Publication of CN112052551A publication Critical patent/CN112052551A/en
Application granted granted Critical
Publication of CN112052551B publication Critical patent/CN112052551B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F04—POSITIVE - DISPLACEMENT MACHINES FOR LIQUIDS; PUMPS FOR LIQUIDS OR ELASTIC FLUIDS
    • F04D—NON-POSITIVE-DISPLACEMENT PUMPS
    • F04D27/00—Control, e.g. regulation, of pumps, pumping installations or pumping systems specially adapted for elastic fluids
    • F04D27/001—Testing thereof; Determination or simulation of flow characteristics; Stall or surge detection, e.g. condition monitoring
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E10/00—Energy generation through renewable energy sources
    • Y02E10/70—Wind energy
    • Y02E10/72—Wind turbines with rotation axis in wind direction

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Mechanical Engineering (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention provides a method and a system for identifying a fan surge operation fault. The identification method comprises the following steps: firstly, acquiring an on-site equipment acoustic signal of fan equipment in the working process; preprocessing the field device acoustic signals to obtain two-dimensional time-frequency data of the field device acoustic signals; then, identifying the two-dimensional time-frequency data by adopting a CNN network to obtain a first identification result; identifying the two-dimensional time-frequency data by adopting an LSTM network to obtain a second identification result; and finally, based on a D-S evidence theory, carrying out information fusion on the first identification result and the second identification result to obtain the identification result of the fan surge operation fault. According to the invention, the fan surge fault diagnosis is carried out by utilizing the field device acoustic signals, a sensor is not required to be installed on the fan, the fault occurrence probability diagnosis is carried out on the surge acoustic signal characteristics by adopting a convolutional neural network and a long-short-time memory network, the fusion diagnosis of two results of a decision layer is carried out by using a D-S evidence theory, and the diagnosis accuracy is improved.

Description

Fan surge operation fault identification method and system
Technical Field
The invention relates to the technical field of equipment fault detection, in particular to a method and a system for identifying a fan surge operation fault.
Background
Along with the continuous expansion of the power plant unit, the requirements of various power equipment for stable operation are increased while the load requirements of the power plant are met. The fan equipment such as a feeding fan, a draught fan, a primary fan and the like bears the pressure of a control hearth, ensures the normal operation of a boiler, and is important auxiliary equipment for the high-efficiency operation of a power plant. When the stability of fan equipment is low, or the valve opening is greatly fluctuated to cause unmatched running speed of the fan and rotary air flow, the fan is easy to surge and malfunction, and the fan works in an unstable area to generate pressure and flow pulsation with the characteristics of wide frequency domain and high amplitude, so that periodic convection aerodynamic noise is caused. The surge fault brings great potential safety hazard to the stable operation of the unit, so the method has very important significance for the real-time monitoring of the running state of the fan and the surge fault early warning. The existing surge fault early warning method mainly comprises the steps of 1, collecting a starting signal of a fan, an in-furnace combustion sign, an operating current detection signal, an outlet pressure detection signal and an opening feedback signal of an outlet regulating valve through a fan outlet regulating valve and a pressure detection device, and sending the signals to a computer control system for calculation and analysis to identify whether the fan has a surge phenomenon; 2. spectral subtraction denoising is carried out on the preprocessed voice signals, a Mel frequency cepstrum coefficient (Mel Frequency Cepstral Coefficient) and data stacking are adopted to form the input of a 3D convolutional neural network, and voiceprint recognition is carried out by utilizing the convolutional neural network (ConvolutionalNeural Network, CNN); 3. and processing the running sound of the automobile engine based on time-frequency two-dimensional processing, judging faults according to AlexNet, and simultaneously identifying the running time state of the engine by using LSTM to assist in more accurate fault judgment. However, the prior art has the following technical defects: 1. when surge occurs, the fan has the characteristics of low load working condition, severe vibration and the like, and most current power plants use the mode of information sensors such as pressure, valve opening, vibration and the like as a judging basis for judging whether surge occurs or not. This approach requires multiple sensors to monitor the device in real time, increasing the cost of the device. Meanwhile, the installation requirements of the sensors are different, the placement positions are also different, and the difficulty of installation and maintenance of the sensors is increased. 2. The existing equipment fault detection and early warning method mostly utilizes deep knowledge of fault characteristics, and the fault characteristics are identified and classified based on the traditional information processing method, so that the method needs higher equipment understanding and operation experience, has poor generalization, and is difficult to identify the fault type or has poor identification rate when other faults occur in equipment or various faults are mixed and appear. 3. The equipment fault identification method has higher requirements on accuracy, and a fault monitoring and early warning algorithm is required to have low rejection rate and low false recognition rate. The signal of equipment surge often shows periodic variation, and the fault characteristic at the early stage of surge occurrence is similar to the running characteristic of the fan for regulating down load, and the requirement of accurate fault early warning cannot be met by carrying out fault diagnosis only from the static angle on the one hand, so that the accuracy is low.
Disclosure of Invention
The invention aims to provide a method and a system for identifying a fan surge operation fault, so as to improve the accuracy of identification on the basis of no need of installing a sensor on fan equipment.
In order to achieve the above object, the present invention provides the following solutions:
a method for identifying a fan surge operation fault, the identification method comprising the steps of:
acquiring an on-site equipment acoustic signal of fan equipment in the working process;
preprocessing the field device acoustic signal to obtain two-dimensional time-frequency data of the field device acoustic signal;
identifying the two-dimensional time-frequency data by adopting a CNN network to obtain a first identification result;
identifying the two-dimensional time-frequency data by adopting an LSTM network to obtain a second identification result;
and based on a D-S evidence theory, carrying out information fusion on the first identification result and the second identification result to obtain the identification result of the fan surge operation fault.
Optionally, the preprocessing the field device acoustic signal to obtain two-dimensional time-frequency data of the field device acoustic signal specifically includes:
carrying out Hamming window filtering and discrete Fourier transform on the field device acoustic signals to obtain frequency domain data of the field device acoustic signals;
And establishing two-dimensional time-frequency data of the field device acoustic signal by taking the time domain data of the field device acoustic signal as one dimension and taking the frequency domain data as the other dimension.
Optionally, the identifying the two-dimensional time-frequency data by using a CNN network to obtain a first identification result specifically includes:
respectively acquiring a normal sound signal when the fan equipment normally operates and a fault sound signal when the fan equipment initially operates in surge, and constructing a normal time frequency data set and a fault time frequency data set;
training a CNN model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained CNN model;
and inputting the two-dimensional time-frequency data into the trained CNN model to obtain the output of the trained CNN model as a first recognition result.
Optionally, training the CNN model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained CNN model, which specifically includes:
initializing a convolution kernel weight and an additive bias vector of each layer of the CNN model;
respectively inputting the normal time-frequency data in the normal time-frequency data set and the fault time-frequency data in the fault time-frequency data set into a first convolution layer of a CNN model, and obtaining an output result of the first convolution layer;
Inputting the output result of the first convolution layer into a first pooling layer of the CNN model to obtain the output result of the first pooling layer;
inputting the output result of the first pooling layer into a second convolution layer of a CNN model to obtain the output result of the second convolution layer;
inputting the output result of the second convolution layer into a second pooling layer of the CNN model, and obtaining the output result of the second pooling layer;
inputting the output result of the second pooling layer into a full-connection layer of the CNN model to obtain the output result of the CNN model;
calculating the error between the output result of the CNN model and the target output result to obtain a CNN prediction error;
judging whether the CNN prediction error is smaller than a first error threshold value or not, and obtaining a first judgment result;
if the first judgment result indicates that the CNN prediction error is not smaller than a first error threshold, judging whether the CNN training times are smaller than a first training times threshold or not, and obtaining a second judgment result;
if the second judgment result indicates that the number of CNN training times is smaller than a first training time threshold, updating the convolution kernel weight and the additive offset vector of each layer of the CNN model, and returning to the step of inputting the normal time-frequency data in the normal time-frequency data set and the fault time-frequency data in the fault time-frequency data set into the first convolution layer of the CNN model respectively to obtain an output result of the first convolution layer;
And if the first judgment result indicates that the CNN prediction error is smaller than a first error threshold or the second judgment result indicates that the CNN training frequency is not smaller than a first training frequency threshold, outputting a trained CNN model.
Optionally, the identifying the two-dimensional time-frequency data by using an LSTM network to obtain a second identification result specifically includes:
respectively acquiring a normal sound signal when the fan equipment normally operates and a fault sound signal when the fan equipment initially operates in surge, and constructing a normal time frequency data set and a fault time frequency data set;
training an LSTM model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained LSTM model;
and inputting the two-dimensional time-frequency data into the trained LSTM model to obtain the output of the trained LSTM model as a second recognition result.
Optionally, training the LSTM model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained LSTM model, which specifically includes:
initializing weight values and additive bias vectors of all layers of an LSTM model;
respectively inputting the normal time-frequency data set and the fault time-frequency data set into an input layer of the LSTM model to obtain output of the input layer;
Inputting the output of the input layer into a first hidden layer of an LSTM model, and obtaining the output of the first hidden layer;
inputting the output of the first hidden layer into a second hidden layer of the LSTM model to obtain the output of the second hidden layer;
inputting the output of the second hidden layer into a full-connection layer of the LSTM model to obtain an output result of the LSTM model;
calculating the error between the output result of the LSTM model and the target output result to obtain an LSTM prediction error;
judging whether the LSTM prediction error is smaller than a second error threshold value or not, and obtaining a third judgment result;
if the third judgment result indicates that the LSTM prediction error is not smaller than a second error threshold, judging whether the LSTM training frequency is smaller than the second training frequency threshold or not, and obtaining a fourth judgment result;
if the fourth judgment result shows that the LSTM training frequency is smaller than the second training frequency threshold, updating the weight and the additive bias vector of each layer of the LSTM model, and returning to the step of inputting the normal time-frequency data set and the fault time-frequency data set into the input layer of the LSTM model respectively to obtain the output of the input layer;
and if the third judgment result indicates that the LSTM prediction error is smaller than a second error threshold or the fourth judgment result indicates that the LSTM training frequency is not smaller than a second training frequency threshold, outputting the trained LSTM model.
A fan surge operation fault identification system, the identification system comprising:
the field device acoustic signal acquisition module is used for acquiring a field device acoustic signal of the fan device in the working process;
the preprocessing module is used for preprocessing the field device acoustic signals to obtain two-dimensional time-frequency data of the field device acoustic signals;
the CNN network identification module is used for identifying the two-dimensional time-frequency data by adopting a CNN network to obtain a first identification result;
the LSTM network identification module is used for identifying the two-dimensional time-frequency data by adopting an LSTM network to obtain a second identification result;
and the recognition result fusion module is used for carrying out information fusion on the first recognition result and the second recognition result based on the D-S evidence theory to obtain the recognition result of the fan surge operation fault.
Optionally, the preprocessing module specifically includes:
the filtering and Fourier transforming sub-module is used for carrying out Hamming window filtering and discrete Fourier transformation on the field device sound signals to obtain frequency domain data of the field device sound signals;
and the two-dimensional time-frequency data establishing sub-module is used for establishing the two-dimensional time-frequency data of the field device acoustic signal by taking the time-domain data of the field device acoustic signal as one dimension and taking the frequency-domain data as the other dimension.
Optionally, the CNN network identification module specifically includes:
the sample data acquisition sub-module is used for respectively acquiring a normal sound signal when the fan equipment operates normally and a fault sound signal when the fan equipment operates in the early surge stage, and constructing a normal time frequency data set and a fault time frequency data set;
the CNN model training sub-module is used for training the CNN model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained CNN model;
and the CNN network identification sub-module is used for inputting the two-dimensional time-frequency data into the trained CNN model to obtain the output of the trained CNN model as a first identification result.
Optionally, the CNN model training submodule specifically includes:
the initialization unit is used for initializing the convolution kernel weight and the additive offset vector of each layer of the CNN model;
the first convolution layer training unit is used for respectively inputting the normal time-frequency data in the normal time-frequency data set and the fault time-frequency data in the fault time-frequency data set into a first convolution layer of a CNN model to obtain an output result of the first convolution layer;
the first pooling layer training unit is used for inputting the output result of the first convolution layer into the first pooling layer of the CNN model to obtain the output result of the first pooling layer;
The second convolution layer training unit is used for inputting the output result of the first pooling layer into a second convolution layer of the CNN model to obtain the output result of the second convolution layer;
the second pooling layer training unit is used for inputting the output result of the second convolution layer into a second pooling layer of the CNN model to obtain the output result of the second pooling layer;
the full-connection layer training unit is used for inputting the output result of the second pooling layer into the full-connection layer of the CNN model to obtain the output result of the CNN model;
the prediction error calculation unit is used for calculating errors of the output result of the CNN model and the target output result to obtain CNN prediction errors;
the first judging unit is used for judging whether the CNN prediction error is smaller than a first error threshold value or not to obtain a first judging result;
the second judging unit is used for judging whether the CNN training times are smaller than a first training times threshold value or not if the first judging result indicates that the CNN prediction error is not smaller than the first error threshold value, so as to obtain a second judging result;
a parameter updating unit, configured to update a convolution kernel weight and an additive bias vector of each layer of the CNN model if the second judgment result indicates that the number of CNN training times is smaller than a first training time threshold, and return to the step of inputting the normal time-frequency data in the normal time-frequency data set and the failure time-frequency data in the failure time-frequency data set into a first convolution layer of the CNN model respectively, to obtain an output result of the first convolution layer;
And the trained CNN model output unit is used for outputting the trained CNN model if the first judgment result indicates that the CNN prediction error is smaller than a first error threshold value or the second judgment result indicates that the CNN training frequency is not smaller than a first training frequency threshold value.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides a method and a system for identifying a fan surge operation fault. The identification method comprises the following steps: firstly, acquiring an on-site equipment acoustic signal of fan equipment in the working process; preprocessing the field device acoustic signals to obtain two-dimensional time-frequency data of the field device acoustic signals; then, identifying the two-dimensional time-frequency data by adopting a CNN network to obtain a first identification result; identifying the two-dimensional time-frequency data by adopting an LSTM network to obtain a second identification result; and finally, based on a D-S evidence theory, carrying out information fusion on the first identification result and the second identification result to obtain the identification result of the fan surge operation fault. According to the invention, the fan surge fault diagnosis is carried out by utilizing the field device acoustic signals, a sensor is not required to be installed on the fan, the fault occurrence probability diagnosis is carried out on the surge acoustic signal characteristics by adopting a convolutional neural network (Convolutional Neural Network, CNN) and a Long Short-term memory network (LSTM) (Long Short-term Memory Network, LSTM), and the fusion diagnosis of two results of a decision layer is carried out by using a D-S evidence theory, so that the diagnosis accuracy is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for identifying a fan surge operation fault provided by the invention;
fig. 2 is a flowchart of CNN network training provided in the present invention;
FIG. 3 is a flow chart of LSTM network training provided by the present invention;
fig. 4 is a block diagram of the storage speed of the LSTM network provided by the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention aims to provide a method and a system for identifying a fan surge operation fault, so as to improve the accuracy of identification on the basis of no need of installing a sensor on fan equipment.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
As shown in fig. 1, the invention provides a method for identifying a fan surge operation fault, which comprises the following steps:
and acquiring an acoustic signal of the field device of the fan device in the working process.
Preprocessing the field device acoustic signal to obtain two-dimensional time-frequency data of the field device acoustic signal; the method specifically comprises the following steps:
carrying out Hamming window filtering and discrete Fourier transform on the field device acoustic signals to obtain frequency domain data of the field device acoustic signals; and establishing two-dimensional time-frequency data of the field device acoustic signal by taking the time domain data of the field device acoustic signal as one dimension and taking the frequency domain data as the other dimension.
Identifying the two-dimensional time-frequency data by adopting a CNN network to obtain a first identification result; the method specifically comprises the following steps:
respectively acquiring a normal sound signal when the fan equipment normally operates and a fault sound signal when the fan equipment initially operates in surge, and constructing a normal time frequency data set and a fault time frequency data set; training a CNN model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained CNN model; and inputting the two-dimensional time-frequency data into the trained CNN model to obtain the output of the trained CNN model as a first recognition result.
Identifying the two-dimensional time-frequency data by adopting an LSTM network to obtain a second identification result; the method specifically comprises the following steps:
respectively acquiring a normal sound signal when the fan equipment normally operates and a fault sound signal when the fan equipment initially operates in surge, and constructing a normal time frequency data set and a fault time frequency data set; training an LSTM model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained LSTM model; and inputting the two-dimensional time-frequency data into the trained LSTM model to obtain the output of the trained LSTM model as a second recognition result.
Based on a D-S evidence theory, carrying out information fusion on the first identification result and the second identification result to obtain an identification result of fan surge operation faults; the fusion mode is shown as follows:
Figure BDA0002247456800000081
Figure BDA0002247456800000082
Figure BDA0002247456800000083
wherein P (normal) is the fusion probability result of normal operation of the fan, P (fault) is the fusion probability result of surging operation of the fan, K is a normalization constant, and P A (normal) is the CNN recognition probability of the normal operation of the fan, P A (fault) identifying probability, P for CNN of fan surge operation B (normal) is the LSTM identification probability of the normal operation of the fan, P B (fault) identifies the probability for LSTM of fan surge operation.
The method comprises the steps of respectively collecting a normal sound signal when fan equipment normally operates and a fault sound signal when the fan equipment initially operates in surge, and constructing a normal time frequency data set and a fault time frequency data set, and specifically comprises the following steps:
1-1, acquiring acoustic signals of fan equipment by using a sound pick-up to obtain two digital sample signals with fixed duration: a normal sound signal (hereinafter referred to as "normal signal") of normal operation of the fan and a normal sound signal (hereinafter referred to as "fault signal") of an early stage of surge of the fan;
1-2 establish a device state data set from a plurality of normal signals and fault signals, the fault signals and normal signals in the device state data set being collectively referred to as sample signals.
Step 2: acquiring two-dimensional time-frequency data
And carrying out short-time Fourier transform on each sample signal in the data set to obtain a plurality of frequency spectrums of each signal, and combining the frequency spectrums with the time domain spectrums to obtain two-dimensional time frequency data serving as input characteristics of the identification network. The method comprises the following specific steps:
2-1 performs an initial N-point hamming window filtering on a sample signal in the device state dataset having a total length of M points, where N < M and the hamming window function is:
Figure BDA0002247456800000084
where n is each signal point within each frame of the audio signal.
2-2 performing discrete Fourier transform on N points in the window to obtain frequency domain data of the sample signal at the moment;
2-3, moving N points of the window to the right, repeating the step 2-1 and the step 2-2, and if the length of the moved signal is less than the N points, filling zero until all the sample signals of the M points are processed. Carrying out Hamming window filtering and Fourier transformation on the next moment of the sample signal until the frequency domain data acquisition of all moments of the sample signal is completed;
2-4 for the sample signal, constructing a two-dimensional time-frequency data as a signal feature by taking time-domain data as one dimension and corresponding frequency-domain data as another dimension;
2-5 repeating the steps 2-1 to 2-4 to obtain a normal time-frequency data set of normal signals in the data set and a fault time-frequency signal data set of fault signals.
As shown in fig. 2, training the CNN model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained CNN model, which specifically includes:
initializing a convolution kernel weight and an additive bias vector of each layer of the CNN model; specifically, the convolution kernel weights are initialized with random values and the additive bias vectors are initialized with zero values. Respectively inputting the normal time-frequency data in the normal time-frequency data set and the fault time-frequency data in the fault time-frequency data set into a first convolution layer of a CNN model, and obtaining an output result of the first convolution layer; inputting the output result of the first convolution layer into a first pooling layer of the CNN model to obtain the output result of the first pooling layer; inputting the output result of the first pooling layer into a second convolution layer of a CNN model to obtain the output result of the second convolution layer; inputting the output result of the second convolution layer into a second pooling layer of the CNN model, and obtaining the output result of the second pooling layer; inputting the output result of the second pooling layer into a full-connection layer of the CNN model to obtain the output result of the CNN model; calculating the error of the output result of the CNN model and the target output result (the target output result is the actual fault category of the training sample, for example, A, B, C fault categories are included, the actual fault category is A, the target output result is 1,0, namely, the probability that the fault category is A is 1, the probability that the fault category is B and the probability that the fault category is C is 0, and the output result of the CNN model is the probability that the state of the training sample belongs to each fault type), so as to obtain CNN prediction error; judging whether the CNN prediction error is smaller than a first error threshold value or not, and obtaining a first judgment result; if the first judgment result indicates that the CNN prediction error is not smaller than a first error threshold, judging whether the CNN training times are smaller than a first training times threshold or not, and obtaining a second judgment result; if the second judgment result indicates that the number of CNN training times is smaller than a first training time threshold, updating the convolution kernel weight and the additive offset vector of each layer of the CNN model, and returning to the step of inputting the normal time-frequency data in the normal time-frequency data set and the fault time-frequency data in the fault time-frequency data set into the first convolution layer of the CNN model respectively to obtain an output result of the first convolution layer; and if the first judgment result indicates that the CNN prediction error is smaller than a first error threshold or the second judgment result indicates that the CNN training frequency is not smaller than a first training frequency threshold, outputting a trained CNN model.
The method comprises the following specific steps:
3-1 initializing all parameters of a convolution network, wherein the convolution kernel weight is initialized by adopting a random value, and the additive bias vector is initialized by adopting a zero value;
3-2 inputting a normal time frequency data set and a fault time frequency data set to perform characteristic training of a convolution layer, wherein the formula is as follows:
X l ＝A(X l-1 ×W l +b l )
wherein X is l For convolutional layer output, A is the activation function, X l-1 To input a local area of data, W l B is convolution kernel weight l Is an additive bias vector;
after the convolution layer output of the whole two-dimensional time-frequency data is completed, the output result is input into a pooling layer for downsampling, and the pooling layer operation is shown as a formula: x is X l ＝S(X l-1 ) Wherein X is l For pooling layer output, S is the maximum pooling downsampling rule, X l-1 And (3) the data is a local area of the input data until the pooling operation of the convolution output graph data is completed.
3-3 repeating the step 3-2 to finish the second convolution and pooling operation;
3-4 inputting the result of step 3-3 into the full connection layer, and obtaining the output value y through the ReLU activation function and the Softmax function d The output result of the CNN model;
3-5 obtaining the error between the output value and the target value y (target output result) by using the Euclidean distance
Figure BDA0002247456800000101
Figure BDA0002247456800000102
/>
3-6, judging whether the error is smaller than an expected value (a first error threshold value), if so, transmitting the error back to the network, sequentially obtaining the errors of all layers, updating the weight and the additive bias vector of each layer of the CNN network, and repeating the steps 3-2 to 3-6 for the next network training; and if the error is smaller than the expected value or the CNN network training frequency reaches the first training frequency threshold value, completing network training to obtain the trained CNN network. The method for updating the convolution kernel weight and the additive offset vector of each layer of the CNN model comprises the following steps: and reversely transmitting the errors along each layer of the CNN, calculating the errors of parameters such as the convolution kernel weight and the additive offset vector of each layer, obtaining the modified value of each parameter through gradient descent, and obtaining the updated parameter by making difference with the original value.
As shown in fig. 3, the training the LSTM model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained LSTM model specifically includes:
initializing weight values and additive bias vectors of all layers of an LSTM model; respectively inputting the normal time-frequency data set and the fault time-frequency data set into an input layer of the LSTM model to obtain output of the input layer; inputting the output of the input layer into a first hidden layer of an LSTM model, and obtaining the output of the first hidden layer; inputting the output of the first hidden layer into a second hidden layer of the LSTM model to obtain the output of the second hidden layer; inputting the output of the second hidden layer into a full-connection layer of the LSTM model to obtain an output result of the LSTM model; calculating the error between the output result of the LSTM model and the target output result to obtain an LSTM prediction error; judging whether the LSTM prediction error is smaller than a second error threshold value or not, and obtaining a third judgment result; if the third judgment result indicates that the LSTM prediction error is not smaller than a second error threshold, judging whether the LSTM training frequency is smaller than the second training frequency threshold or not, and obtaining a fourth judgment result; if the fourth judgment result shows that the LSTM training frequency is smaller than the second training frequency threshold, updating the weight and the additive bias vector of each layer of the LSTM model, and returning to the step of inputting the normal time-frequency data set and the fault time-frequency data set into the input layer of the LSTM model respectively to obtain the output of the input layer; and if the third judgment result indicates that the LSTM prediction error is smaller than a second error threshold or the fourth judgment result indicates that the LSTM training frequency is not smaller than a second training frequency threshold, outputting the trained LSTM model. The method comprises the following specific steps:
Initializing each parameter weight and bias of an LSTM network;
4-2 inputting the normal time-frequency data set and the fault time-frequency data set into an input layer of the LSTM network, wherein all frequency-domain data of the same time-domain dimension of a certain time-frequency data represents an input node, and the time-frequency data can be transmitted to the next layer as characteristics;
4-3 the time-frequency data features are processed in two hidden layers. Each hidden layer is provided with a plurality of time step processing modules, each time step module is provided with time sequence operation capability, the time step modules corresponding to the two hidden layers are provided with layer sequence operation capability, as shown in fig. 4, each time step module is provided with a storage block structure comprising an input door, a forgetting door and an output door, and the calculation relation of each time step is shown in the following formula:
Figure BDA0002247456800000111
Figure BDA0002247456800000112
Figure BDA0002247456800000113
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure BDA0002247456800000114
for input gate output, +.>
Figure BDA0002247456800000115
Output for forgetting gate, ++>
Figure BDA0002247456800000116
For outputting gate output, +.>
Figure BDA0002247456800000117
Memory cell parameters for the current time step, +.>
Figure BDA0002247456800000118
For the previous time step memory cell parameter, +.>
Figure BDA0002247456800000119
For the local feature input area corresponding to the time step of the last hidden layer +.>
Figure BDA00022474568000001110
For a time-step local feature input area on the same hidden layer, sign represents a nonlinear Sigmoid activation function, tanh represents a Tanh activation function, +.>
Figure BDA0002247456800000121
Representation->
Figure BDA0002247456800000122
And->
Figure BDA0002247456800000123
Is activated by tanh, W l Weight matrix representing current hidden layer, b l An additive vector bias representing the current hidden layer, < ->
Figure BDA0002247456800000124
Is the output of the current time step.
Transmitting the output to the next layer after completing the hidden layer calculation;
4-4 inputting the output result of the step 4-3 into the full connection layer, and obtaining an output value y 'through a ReLU activation function and a Softmax function' d I.e. LSTM network output results;
4-5 obtaining the error between the output value and the target value y' (target output result) by using the Euclidean distance
Figure BDA0002247456800000125
Figure BDA0002247456800000126
4-6 judging whether the error is smaller than an expected value (a second error threshold value), if so, reversely calculating the error term value of each neuron, wherein error transmission needs to be carried out along two directions, namely, the error term at each moment before calculation is carried out along the reverse propagation of time, namely, from the current moment t; the other is to propagate the error item to the previous hidden layer, calculate the parameter gradient and update the parameter value according to the corresponding error item, repeat steps 4-2 to 4-6, and carry out the next network training; and if the error is smaller than the expected value or the network training frequency reaches the maximum value, completing the network training to obtain the trained LSTM network. The specific steps for updating the weight and the additive bias vector of each layer of the LSTM model comprise the following steps: the specific error transfer mode is consistent with the error transfer mode in the convolution kernel weight of each layer of the CNN model and the updating mode of the additive offset vector, and is characterized in that the LSTM network carries out error transfer along two directions, one is counter-propagation along time, namely, from the current t moment, calculating an error item at each moment before; the other is to propagate the error term to the previous hidden layer, calculate the parameter gradient and update the parameter value according to the corresponding error term.
The invention also provides a fan surge operation fault identification system, which comprises:
the field device acoustic signal acquisition module is used for acquiring a field device acoustic signal of the fan device in the working process; the method specifically comprises the following steps:
the filtering and Fourier transforming sub-module is used for carrying out Hamming window filtering and discrete Fourier transformation on the field device sound signals to obtain frequency domain data of the field device sound signals; and the two-dimensional time-frequency data establishing sub-module is used for establishing the two-dimensional time-frequency data of the field device acoustic signal by taking the time-domain data of the field device acoustic signal as one dimension and taking the frequency-domain data as the other dimension.
The preprocessing module is used for preprocessing the field device acoustic signals to obtain two-dimensional time-frequency data of the field device acoustic signals;
the CNN network identification module is used for identifying the two-dimensional time-frequency data by adopting a CNN network to obtain a first identification result; the method specifically comprises the following steps:
the sample data acquisition sub-module is used for respectively acquiring a normal sound signal when the fan equipment operates normally and a fault sound signal when the fan equipment operates in the early surge stage, and constructing a normal time frequency data set and a fault time frequency data set; the CNN model training sub-module is used for training the CNN model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained CNN model; and the CNN network identification sub-module is used for inputting the two-dimensional time-frequency data into the trained CNN model to obtain the output of the trained CNN model as a first identification result.
The CNN model training sub-module specifically comprises: the initialization unit is used for initializing the convolution kernel weight and the additive offset vector of each layer of the CNN model; the first convolution layer training unit is used for respectively inputting the normal time-frequency data in the normal time-frequency data set and the fault time-frequency data in the fault time-frequency data set into a first convolution layer of a CNN model to obtain an output result of the first convolution layer; the first pooling layer training unit is used for inputting the output result of the first convolution layer into the first pooling layer of the CNN model to obtain the output result of the first pooling layer; the second convolution layer training unit is used for inputting the output result of the first pooling layer into a second convolution layer of the CNN model to obtain the output result of the second convolution layer; the second pooling layer training unit is used for inputting the output result of the second convolution layer into a second pooling layer of the CNN model to obtain the output result of the second pooling layer; the full-connection layer training unit is used for inputting the output result of the second pooling layer into the full-connection layer of the CNN model to obtain the output result of the CNN model; the prediction error calculation unit is used for calculating errors of the output result of the CNN model and the target output result to obtain CNN prediction errors; the first judging unit is used for judging whether the CNN prediction error is smaller than a first error threshold value or not to obtain a first judging result; the second judging unit is used for judging whether the CNN training times are smaller than a first training times threshold value or not if the first judging result indicates that the CNN prediction error is not smaller than the first error threshold value, so as to obtain a second judging result; a parameter updating unit, configured to update a convolution kernel weight and an additive bias vector of each layer of the CNN model if the second judgment result indicates that the number of CNN training times is smaller than a first training time threshold, and return to the step of inputting the normal time-frequency data in the normal time-frequency data set and the failure time-frequency data in the failure time-frequency data set into a first convolution layer of the CNN model respectively, to obtain an output result of the first convolution layer; and the trained CNN model output unit is used for outputting the trained CNN model if the first judgment result indicates that the CNN prediction error is smaller than a first error threshold value or the second judgment result indicates that the CNN training frequency is not smaller than a first training frequency threshold value.
The LSTM network identification module is used for identifying the two-dimensional time-frequency data by adopting an LSTM network to obtain a second identification result;
and the recognition result fusion module is used for carrying out information fusion on the first recognition result and the second recognition result based on the D-S evidence theory to obtain the recognition result of the fan surge operation fault.
The invention has the advantages that the cost and inconvenience brought by the installation and maintenance of multiple sensors are reduced by combining the acoustic processing technology, and the algorithm for judging the single information is easier to realize. The traditional audio information processing technology and the deep learning method are combined, so that the fault recognition rate is greatly improved. The recognition results of the CNN network and the LSTM network are subjected to decision fusion through a D-S evidence theory, and fault recognition is performed from the two aspects of static judgment of the CNN network and dynamic memory of the LSTM network, so that the accuracy of information judgment is improved, and the fault early warning and prediction capability is enhanced.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The principles and embodiments of the present invention have been described herein with reference to specific examples, which are intended to be only illustrative of the methods and concepts underlying the invention, and not all examples are intended to be within the scope of the invention as defined by the appended claims.

Claims (5)

1. A method for identifying a fan surge operation fault, the method comprising the steps of:
acquiring an on-site equipment acoustic signal of fan equipment in the working process;
preprocessing the field device acoustic signal to obtain two-dimensional time-frequency data of the field device acoustic signal, wherein the method specifically comprises the following steps: carrying out Hamming window filtering and discrete Fourier transform on the field device acoustic signals to obtain frequency domain data of the field device acoustic signals; establishing two-dimensional time-frequency data of the field device acoustic signal by taking the time-domain data of the field device acoustic signal as one dimension and the frequency-domain data as the other dimension;
identifying the two-dimensional time-frequency data by adopting a CNN network to obtain a first identification result, wherein the method specifically comprises the following steps: respectively acquiring a normal sound signal when the fan equipment normally operates and a fault sound signal when the fan equipment initially operates in surge, and constructing a normal time frequency data set and a fault time frequency data set; training a CNN model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained CNN model; inputting the two-dimensional time-frequency data into the trained CNN model to obtain the output of the trained CNN model as a first recognition result;
The LSTM network is adopted to identify the two-dimensional time-frequency data, and a second identification result is obtained, which comprises the following steps: respectively acquiring a normal sound signal when the fan equipment normally operates and a fault sound signal when the fan equipment initially operates in surge, and constructing a normal time frequency data set and a fault time frequency data set; training an LSTM model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained LSTM model; inputting the two-dimensional time-frequency data into the trained LSTM model to obtain the output of the trained LSTM model as a second recognition result;
based on a D-S evidence theory, carrying out information fusion on the first identification result and the second identification result to obtain an identification result of fan surge operation faults; the fusion mode is shown as follows:
Figure FDA0004145996030000011
Figure FDA0004145996030000012
Figure FDA0004145996030000013
wherein P (normal) is the fusion probability result of normal operation of the fan, P (fault) is the fusion probability result of surging operation of the fan, K is a normalization constant, and P A (normal) is the CNN recognition probability of the normal operation of the fan, P A (fault) identifying probability, P for CNN of fan surge operation B (normal) is the LSTM identification probability of the normal operation of the fan, P B (fault) identifies the probability for LSTM of fan surge operation.
2. The fan surge operation fault identification method according to claim 1, wherein training the CNN model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained CNN model specifically comprises:
initializing a convolution kernel weight and an additive bias vector of each layer of the CNN model;
respectively inputting the normal time-frequency data in the normal time-frequency data set and the fault time-frequency data in the fault time-frequency data set into a first convolution layer of a CNN model, and obtaining an output result of the first convolution layer;
inputting the output result of the first convolution layer into a first pooling layer of the CNN model to obtain the output result of the first pooling layer;
inputting the output result of the first pooling layer into a second convolution layer of a CNN model to obtain the output result of the second convolution layer;
inputting the output result of the second convolution layer into a second pooling layer of the CNN model, and obtaining the output result of the second pooling layer;
inputting the output result of the second pooling layer into a full-connection layer of the CNN model to obtain the output result of the CNN model;
calculating the error between the output result of the CNN model and the target output result to obtain a CNN prediction error;
judging whether the CNN prediction error is smaller than a first error threshold value or not, and obtaining a first judgment result;
If the first judgment result indicates that the CNN prediction error is not smaller than a first error threshold, judging whether the CNN training times are smaller than a first training times threshold or not, and obtaining a second judgment result;
if the second judgment result indicates that the number of CNN training times is smaller than a first training time threshold, updating the convolution kernel weight and the additive offset vector of each layer of the CNN model, and returning to the step of inputting the normal time-frequency data in the normal time-frequency data set and the fault time-frequency data in the fault time-frequency data set into the first convolution layer of the CNN model respectively to obtain an output result of the first convolution layer;
and if the first judgment result indicates that the CNN prediction error is smaller than a first error threshold or the second judgment result indicates that the CNN training frequency is not smaller than a first training frequency threshold, outputting a trained CNN model.
3. The method for identifying a fan surge operation fault according to claim 1, wherein training an LSTM model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained LSTM model specifically comprises:
initializing weight values and additive bias vectors of all layers of an LSTM model;
respectively inputting the normal time-frequency data set and the fault time-frequency data set into an input layer of the LSTM model to obtain output of the input layer;
Inputting the output of the input layer into a first hidden layer of an LSTM model, and obtaining the output of the first hidden layer;
inputting the output of the first hidden layer into a second hidden layer of the LSTM model to obtain the output of the second hidden layer;
inputting the output of the second hidden layer into a full-connection layer of the LSTM model to obtain an output result of the LSTM model;
calculating the error between the output result of the LSTM model and the target output result to obtain an LSTM prediction error;
judging whether the LSTM prediction error is smaller than a second error threshold value or not, and obtaining a third judgment result;
if the third judgment result indicates that the LSTM prediction error is not smaller than a second error threshold, judging whether the LSTM training frequency is smaller than the second training frequency threshold or not, and obtaining a fourth judgment result;
if the fourth judgment result shows that the LSTM training frequency is smaller than the second training frequency threshold, updating the weight and the additive bias vector of each layer of the LSTM model, and returning to the step of inputting the normal time-frequency data set and the fault time-frequency data set into the input layer of the LSTM model respectively to obtain the output of the input layer;
and if the third judgment result indicates that the LSTM prediction error is smaller than a second error threshold or the fourth judgment result indicates that the LSTM training frequency is not smaller than a second training frequency threshold, outputting the trained LSTM model.
4. A fan surge operation fault identification system, the identification system comprising:
the field device acoustic signal acquisition module is used for acquiring a field device acoustic signal of the fan device in the working process;
the preprocessing module is used for preprocessing the field device acoustic signals to obtain two-dimensional time-frequency data of the field device acoustic signals; the preprocessing module specifically comprises: the filtering and Fourier transforming sub-module is used for carrying out Hamming window filtering and discrete Fourier transformation on the field device sound signals to obtain frequency domain data of the field device sound signals; the two-dimensional time-frequency data establishing sub-module is used for establishing two-dimensional time-frequency data of the field device acoustic signal by taking the time-domain data of the field device acoustic signal as one dimension and taking the frequency-domain data as the other dimension;
the CNN network identification module is used for identifying the two-dimensional time-frequency data by adopting a CNN network to obtain a first identification result; the CNN network identification module specifically comprises: the sample data acquisition sub-module is used for respectively acquiring a normal sound signal when the fan equipment operates normally and a fault sound signal when the fan equipment operates in the early surge stage, and constructing a normal time frequency data set and a fault time frequency data set; the CNN model training sub-module is used for training the CNN model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained CNN model; the CNN network identification sub-module is used for inputting the two-dimensional time-frequency data into the trained CNN model, and obtaining the output of the trained CNN model as a first identification result;
The LSTM network identification module is used for identifying the two-dimensional time-frequency data by adopting an LSTM network to obtain a second identification result, and specifically comprises the following steps: respectively acquiring a normal sound signal when the fan equipment normally operates and a fault sound signal when the fan equipment initially operates in surge, and constructing a normal time frequency data set and a fault time frequency data set; training an LSTM model according to the normal time-frequency data set and the fault time-frequency data set to obtain a trained LSTM model; inputting the two-dimensional time-frequency data into the trained LSTM model to obtain the output of the trained LSTM model as a second recognition result;
the recognition result fusion module is used for carrying out information fusion on the first recognition result and the second recognition result based on a D-S evidence theory to obtain a recognition result of the fan surge operation fault; the fusion mode is shown as follows:
Figure FDA0004145996030000041
Figure FDA0004145996030000051
Figure FDA0004145996030000052
wherein P (normal) is the fusion probability result of normal operation of the fan, P (fault) is the fusion probability result of surging operation of the fan, K is a normalization constant, and P A (normal) is the CNN recognition probability of the normal operation of the fan, P A (fault) identifying probability, P for CNN of fan surge operation B (normal) is the LSTM identification probability of the normal operation of the fan, P B (fault) identifies the probability for LSTM of fan surge operation.
5. The fan surge operation fault identification system of claim 4, wherein the CNN model training sub-module specifically comprises:
the initialization unit is used for initializing the convolution kernel weight and the additive offset vector of each layer of the CNN model;
the first convolution layer training unit is used for respectively inputting the normal time-frequency data in the normal time-frequency data set and the fault time-frequency data in the fault time-frequency data set into a first convolution layer of a CNN model to obtain an output result of the first convolution layer;
the first pooling layer training unit is used for inputting the output result of the first convolution layer into the first pooling layer of the CNN model to obtain the output result of the first pooling layer;
the second convolution layer training unit is used for inputting the output result of the first pooling layer into a second convolution layer of the CNN model to obtain the output result of the second convolution layer;
the second pooling layer training unit is used for inputting the output result of the second convolution layer into a second pooling layer of the CNN model to obtain the output result of the second pooling layer;
the full-connection layer training unit is used for inputting the output result of the second pooling layer into the full-connection layer of the CNN model to obtain the output result of the CNN model;
The prediction error calculation unit is used for calculating errors of the output result of the CNN model and the target output result to obtain CNN prediction errors;
the first judging unit is used for judging whether the CNN prediction error is smaller than a first error threshold value or not to obtain a first judging result;
the second judging unit is used for judging whether the CNN training times are smaller than a first training times threshold value or not if the first judging result indicates that the CNN prediction error is not smaller than the first error threshold value, so as to obtain a second judging result;
a parameter updating unit, configured to update a convolution kernel weight and an additive bias vector of each layer of the CNN model if the second judgment result indicates that the number of CNN training times is smaller than a first training time threshold, and return to the step of inputting the normal time-frequency data in the normal time-frequency data set and the failure time-frequency data in the failure time-frequency data set into a first convolution layer of the CNN model respectively, to obtain an output result of the first convolution layer;
and the trained CNN model output unit is used for outputting the trained CNN model if the first judgment result indicates that the CNN prediction error is smaller than a first error threshold value or the second judgment result indicates that the CNN training frequency is not smaller than a first training frequency threshold value.
CN201911021826.5A 2019-10-25 2019-10-25 Fan surge operation fault identification method and system Active CN112052551B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911021826.5A CN112052551B (en) 2019-10-25 2019-10-25 Fan surge operation fault identification method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911021826.5A CN112052551B (en) 2019-10-25 2019-10-25 Fan surge operation fault identification method and system

Publications (2)

Publication Number Publication Date
CN112052551A CN112052551A (en) 2020-12-08
CN112052551B true CN112052551B (en) 2023-05-02

Family

ID=73609742

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911021826.5A Active CN112052551B (en) 2019-10-25 2019-10-25 Fan surge operation fault identification method and system

Country Status (1)

Country Link
CN (1) CN112052551B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113566948A (en) * 2021-07-09 2021-10-29 中煤科工集团沈阳研究院有限公司 Fault audio recognition and diagnosis method for robot coal pulverizer
CN115306754B (en) * 2022-10-12 2023-02-17 中国航发四川燃气涡轮研究院 Axial flow fan aerodynamic instability identification method based on acoustic array
CN116393217B (en) * 2023-02-24 2024-07-09 华能曲阜热电有限公司 Intelligent monitoring method for material level of steel ball coal mill

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019056471A1 (en) * 2017-09-19 2019-03-28 平安科技（深圳）有限公司 Driving model training method, driver recognition method and apparatus, device, and medium
CN110131193A (en) * 2018-02-02 2019-08-16 中国航发商用航空发动机有限责任公司 Aero-engine surge fault monitoring method and system

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2991706C (en) * 2015-07-08 2023-10-31 Sensanna Incorporated Interference immune radio

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019056471A1 (en) * 2017-09-19 2019-03-28 平安科技（深圳）有限公司 Driving model training method, driver recognition method and apparatus, device, and medium
CN110131193A (en) * 2018-02-02 2019-08-16 中国航发商用航空发动机有限责任公司 Aero-engine surge fault monitoring method and system

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
Ruochen Duan 等.Fault Diagnosis of On-Load Tap-Changer in Converter Transformer Based on Time–Frequency Vibration Analysis.《IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS》.2016,第63卷(第6期),第3815-3823页. *
Time amplifying techniques towards atomic time resolution.Science in China(Series E:Technological Sciences).2009,(第12期),第3425-3446页. *
Yongjie Zhai 等.Multiscale Entropy Feature Extraction Method of Running Power Equipment Sound.《Entropy 2020》.2020,第1-18页. *
李粮 ; 田青林 ; 钟明 ; 赖小皇 ; .基于Mallat算法的机载失速喘振辨识装置研制.燃气涡轮试验与研究.2019,(第06期),第49-53页. *
李长征 ; 胡智琦 ; 许思琦 ; .基于D-S证据融合的压气机气动失稳检测.推进技术.2017,(第08期),第195-202页. *
陈敏红.基于时频分析的轴流风机故障诊断实验研究.《中国优秀博硕士学位论文全文数据库（硕士）信息科技II辑》.2006,全文. *

Also Published As

Publication number Publication date
CN112052551A (en) 2020-12-08

Similar Documents

Publication Publication Date Title
CN112052551B (en) Fan surge operation fault identification method and system
CN113567131B (en) Bearing fault diagnosis method based on S transformation and miniature convolutional neural network model
Flah et al. Localization and classification of structural damage using deep learning single-channel signal-based measurement
CN112393934A (en) Wind turbine generator fault diagnosis method based on sparse self-coding and extreme learning machine
CN116776208B (en) Training method of seismic wave classification model, seismic wave selecting method, equipment and medium
CN115859077A (en) Multi-feature fusion motor small sample fault diagnosis method under variable working conditions
US20220269988A1 (en) Abnormality degree calculation system and abnormality degree calculation method
Hu et al. Status detection from spatial-temporal data in pipeline network using data transformation convolutional neural network
CN117809164A (en) Substation equipment fault detection method and system based on multi-mode fusion
CN116012681A (en) Method and system for diagnosing motor faults of pipeline robot based on sound vibration signal fusion
CN115047290A (en) Cable fault discharge sound detection method based on deep learning
CN111539157B (en) Dynamic load time domain identification method based on multilayer perceptron
Du et al. Stacked convolutional LSTM models for prognosis of bearing performance degradation
CN117332324A (en) Pipeline leakage detection method and device, electronic equipment and storage medium
CN114970745B (en) Intelligent security and environment big data system of Internet of things
Čavor et al. Vehicle speed estimation from audio signals using 1d convolutional neural networks
Hao et al. New fusion features convolutional neural network with high generalization ability on rolling bearing fault diagnosis
CN114942480A (en) Ocean station wind speed forecasting method based on information perception attention dynamic cooperative network
KR20230072920A (en) Iterative Learning Method and Apparatus for Unsupervised Anomaly Detection Under Data Contamination in IoT Environment
CN108960406B (en) MEMS gyroscope random error prediction method based on BFO wavelet neural network
Liu et al. A sudden fault detection network based on Time-sensitive gated recurrent units for bearings
CN112560252A (en) Prediction method for residual life of aircraft engine
Kotani et al. Detection of gas leakage sound using modular neural networks for unknown environments
Jung et al. A Diagnostic Technology of Powertrain Parts that Cause Abnormal Noises Using Artificial Intelligence
CN117892188B (en) Fault classification method based on bearing vibration signal relation and graph neural network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant