CN111488935A - Coal mill fault diagnosis method based on neural network unified modeling - Google Patents

Coal mill fault diagnosis method based on neural network unified modeling Download PDF

Info

Publication number
CN111488935A
CN111488935A CN202010290250.9A CN202010290250A CN111488935A CN 111488935 A CN111488935 A CN 111488935A CN 202010290250 A CN202010290250 A CN 202010290250A CN 111488935 A CN111488935 A CN 111488935A
Authority
CN
China
Prior art keywords
fault
sample
data frame
type
level
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.)
Withdrawn
Application number
CN202010290250.9A
Other languages
Chinese (zh)
Inventor
刘加
卢回忆
张卫强
李飞
刘德广
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.)
Beijing Huacong Zhijia Technology Co ltd
Original Assignee
Beijing Huacong Zhijia Technology Co ltd
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 Beijing Huacong Zhijia Technology Co ltd filed Critical Beijing Huacong Zhijia Technology Co ltd
Priority to CN202010290250.9A priority Critical patent/CN111488935A/en
Publication of CN111488935A publication Critical patent/CN111488935A/en
Withdrawn legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • 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
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention relates to a coal mill fault diagnosis method based on neural network unified modeling, and belongs to the technical field of machine fault detection methods and artificial intelligence. Firstly, collecting the rotating speed, the current of a driving motor, the voltage of the driving motor, the coal feeding amount, the primary air pressure, the pressure of a pull rod, the temperature, the vibration and the sound of a coal mill to be detected in the running process, adding sample labels to sample characteristics, and generating a classified and graded fault sample set; randomly extracting samples from the samples in the fault-free state to form a fault-free sample set; the failure sample set and the non-failure sample set constitute a complete machine state sample set. And coding the fault type and the grade according to a binary mode, and splicing the codes of different types and different grades into a code with unified fault detection, classification and grading. The model obtained by training in the method has higher diagnosis accuracy and prediction capability, and can simultaneously diagnose the grading results of various fault types under the condition that various fault types coexist.

Description

Coal mill fault diagnosis method based on neural network unified modeling
Technical Field
The invention relates to a coal mill fault diagnosis method based on neural network unified modeling, and belongs to the technical field of machine fault detection methods and artificial intelligence.
Background
Coal mills are common equipment of thermal power plants, and the failure thereof is one of the most common causes affecting the operation indexes of the power plants and even causing the unplanned shutdown. At present, the conventional coal mill maintenance scheme of a power plant is a planned maintenance scheme adopting a regular maintenance mode. The planned solution does not take into account the actual operating conditions of the machine, and therefore there is the problem that the machine is taken down for service (over-maintenance) in perfect working conditions, while the machine at the fault edge is ignored (under-maintenance). Due to the critical role of coal mills to ensure the productivity of power plants, unscheduled shutdown maintenance will likely lead to significant economic and social consequences. To avoid accidental shutdown of the coal mill, the actual coal mill maintenance solution is a planned maintenance solution that employs "over maintenance".
With the development of information technology and artificial intelligence technology, a predictive maintenance scheme for performing online detection by mounting a sensor on a machine and performing fault diagnosis by using an artificial intelligence model is an important research field of industrial intelligence at present. However, since the artificial intelligence model requires targeted training of different devices, it takes a long time for data accumulation before each monitoring and diagnostic scheme is formally put on line. And because the training samples are completely labeled by human experts, the trained diagnosis model saves a great amount of human labor, can copy the continuous monitoring that the experience of the experts is not tired, but has limited capability on the predictive diagnosis of the potential faults which are difficult to be perceived by human beings. Therefore, the development of the online machine fault monitoring and artificial intelligence diagnosis technology requires breakthrough innovation in the aspects of improving the diagnosis accuracy, improving the prediction capability of diagnosis in advance and the like.
Disclosure of Invention
The invention aims to provide a coal mill fault diagnosis method based on neural network unified modeling so as to improve the intelligence, the practicability and the accuracy of an online coal mill fault diagnosis technology and the predictability of potential faults.
The invention provides a coal mill fault diagnosis method based on neural network unified modeling, which comprises the following steps:
(1) collecting operation state monitoring data of a coal mill to be detected, wherein the operation state monitoring data comprises rotation speed data R, driving motor current I, driving motor voltage E, coal feeding quantity C, primary air pressure W, pull rod pressure P, temperature data T, vibration data V and sound data S of the coal mill to be detected, and R, I, E, C, W, P, T, V and S are time sequence data;
(2) framing the operation state monitoring data acquired in the step (1), setting the duration of a data frame to be tlen, and setting the starting time of the ith data frame to be tiIntercept the time window [ t ]i,ti+tlen]R, I, E, C, W, P, T, V, S in is denoted as R _ ti、I_ti、E_ti、C_ti、W_ti、P_ti、T_ti、V_tiAnd S _ tiAfter the framing processing, R, I, E, C, W, P, T, V, S data are divided into N data frames, which are recorded as:
Figure BDA0002450121880000021
wherein N is the total number of data frames, i is the number of data frames, tiFor the start time of the ith data Frame, Frame _ tiRepresents tiData Frame of time, Frame _ tiFrom tiFrame of rotational speed data at time R _ tiAnd a driving motor current data frame I _ tiAnd a driving motor voltage data frame E _ tiCoal feeding amount data frame C _ tiPrimary wind pressure data frame W _ tiPull rod pressure data frame P _ tiTemperature data frame T _ TiVibration data frame V _ tiAnd a sound data frame S _ tiCombining the components;
(3) for the data Frame _ t obtained in the step (2)iProcessing is carried out to generate a sample Feature set Feature _ Full, and the specific steps are as follows:
(3.1) for Frame _ tiR _ t in (1)i、I_ti、E_ti、C_ti、W_ti、P_ti、T_ti、V_tiAnd S _ tiRespectively carrying out time domain amplitude statistics by using a rotating speed data frame R _ tiObtaining the average value R _ mean _ t of the rotating speediFrom the drive motor current data frame I _ tiObtaining the average value I _ mean _ t of the current of the driving motoriVoltage data frame E _ t of driving motoriObtaining the average value E _ mean _ t of the voltage of the driving motoriFrom the coal input data frame C _ tiObtaining the average value C _ mean _ t of the coal feeding quantityiThe primary wind pressure data frame W _ tiObtaining the primary wind pressure average value W _ mean _ tiFrom the pull rod pressure data frame P _ tiObtaining the average value P _ mean _ t of the pressure of the pull rodiFrom the temperature data frame T _ TiObtaining the temperature average value T _ mean _ TiFrom the vibration data frame V _ tiCalculating to obtain a vibration root mean square V _ rms _ tiVibration variance V _ sigm _ tiVibration skewness V _ skew _ tiAnd the vibration kurtosis V _ kurt _ tiFrom the sound data frame S _ tiCalculating to obtain the root mean square S _ rms _ t of the soundiSound variance S _ sigm _ tiSound skewness S _ skew _ tiAnd sound kurtosis S _ kurt _ tiThe statistics values are spliced into a 15-dimensional vector Vec1_ tiN Frame _ tiThe data frame processing obtains N15-dimensional vectors, which are recorded as:
Figure BDA0002450121880000031
(3.2) for Frame _ tiV _ t in (1)iPerforming fast Fourier transform to obtain a vibration energy spectrum, and performing L subband filtering on the vibration energy spectrum to obtain a L-dimensional vector Vec2_ tiN number of V _ tiThe data frame processing yields N L-dimensional vectors, denoted as:
Figure BDA0002450121880000032
wherein L is the number of Filter subbands, L is a value range of 10-1000, FFT represents fast fourier transform, Filter _ l represents the first subband filtering;
(3.3) for Frame _ tiS _ t in (1)iPerforming fast Fourier transform to obtain a sound energy spectrum, and performing M-subband filtering on the sound energy spectrum to obtain an M-dimensional vector Vec3_ tiN S _ tiProcessing the data frame to obtain N M-dimensional vectors, which are recorded as:
Figure BDA0002450121880000041
wherein M is the number of filtering sub-bands, the value of M is 10-1000, FFT represents fast Fourier transform, Filter _ M represents mth sub-band filtering;
(3.4) mixing Vec1_ t obtained in the steps (3.1) to (3.3)i、Vec2_tiAnd Vec3_ tiSplicing generates a (15+ L + M) -dimensional vector as tiSample characteristic Vec _ t of time instantiAnd is recorded as:
Figure BDA0002450121880000042
(3.5) obtaining N sample characteristics Vec _ t obtained in the step (3.4)iThe set constitutes a sample Feature set Feature _ Full, noted as:
Figure BDA0002450121880000043
(4) training sample set generation, comprising the steps of:
(4.1) dividing the machine status into a fault status and a no fault status, wherein the fault status comprises a plurality of non-classified fault types and a plurality of classified fault type statuses, and dividing the classified fault types into 5 levels according to severity degrees: wherein level 1 indicates in early stage of the fault, level 2 indicates in early stage of the fault, level 3 indicates in middle stage of the fault, level 4 indicates in middle and late stage of the fault, level 5 indicates in late stage of the fault, and the fault duration length vectors of 5 fault levels are represented as [ D1, D2, D3, D4, D5 ]; wherein D1 represents the duration length of progression from level 1 to level 2, D2 represents the duration length of progression from level 2 to level 3, D3 represents the duration length of progression from level 3 to level 4, D4 represents the duration length of progression from level 4 to level 5, D5 represents the duration length from discovery of D5 signature to occurrence of a destructive fault, each classified fault type determines the fault duration length vector as [ D1_ type, D2_ type, D3_ type, D4_ type, D5_ type ], which represents the fault type number in the manner described above;
(4.2) generating a Fault sample Set _ Fault according to a machine Fault marking log obtained from a machine operation maintenance management department from the sample Feature Set Feature _ Full obtained in the step (3.5), wherein the method specifically comprises the following steps:
(4.2.1) extracting a record from the machine fault labeling log, wherein the record content comprises a group of quaternary data in the form of: (τ, type, level, τ)2) Wherein tau is the time when the fault is detected, type is the fault type, level is the fault level, tau is2The moment when the fault is repaired;
(4.2.2) judging the fault type:
if type is not classified fault type, executing steps (4.2.2.1) - (4.2.2.2):
(4.2.2.1) extracting all sample features Vec _ t in a period of time when tau is less than or equal to t < tau 2 from the sample Feature set Feature _ Full obtained in the step (3.5);
(4.2.2.2) adding a Sample label (type) to each Sample feature Vec _ t obtained in the step (4.2.2.1) to generate a labeled fault Sample _ fault ═ (Vec _ t, (type)), wherein Sample _ fault represents a labeled fault Sample, Vec _ t is a Sample feature, (type) is a Sample label, type is a fault type, and "×" is a default item;
if the type is a classified fault type, setting the number of the fault type as itype, and executing the steps (4.2.2.3) - (4.2.2.4):
(4.2.2.3) τ, level, τ obtained according to step (4.2.1)2And the fault duration length vector [ D1_ type, D2_ type, D3_ type, D4_ type, D5_ type ] obtained in the step (4.1)]When the conversion time of the faults in different levels is calculated, taking level as 5 as an example, and level is 4, 3, 2, 1, the processing method is analogized, and the result is as follows:
Figure BDA0002450121880000051
(4.2.2.4) processing each Sample vec _ t in the Sample Feature set Feature _ Full obtained in step (3.5) as follows to generate a labeled fault Sample _ fault:
Figure BDA0002450121880000061
(4.2.3) traversing each record in the machine Fault labeling log, circularly executing the steps (4.2.1) and (4.2.2), and combining all marked Fault samples generated by the steps (4.2.2.2) and (4.2.2.4) into a Set to obtain a Fault sample Set _ Fault;
(4.3) generating a fault-free sample Set _ Normal, comprising the steps of:
(4.3.1) marking the sample Feature corresponding to the Fault sample Set _ Fault generated in the step (4.2.3) as a Fault sample Feature _ Fault, removing the Feature _ Fault from the sample Feature Set Feature _ Full obtained in the step (3.5), marking the residual sample Feature Set as a non-Fault sample Feature Set Feature _ Normal, and meeting a Set operation formula:
Feature_Normal＝Feature_Full-Feature_fault
(4.3.2) randomly extracting a sample Feature vec _ t from the failure-free sample Feature set Feature _ Normal obtained in the step (4.3.1), adding a sample label, and generating a labeled failure-free sample record:
sample _ normal ═ (vec _ t, (no fault type,));
wherein, Sample _ normal is a labeled no-fault Sample, vec _ t is a Sample characteristic, (no-fault type, is a Sample label), "no-fault type" is a machine state type, and "+" is a default term;
(4.3.3) circularly executing the step (4.3.2) to obtain marked Fault-free samples with the number equal to that of the samples of the Fault sample Set _ Fault, and collecting to form a Fault-free sample Set _ Normal;
(4.4) merging the Fault sample Set _ Fault obtained in the step (4.2) with the Fault-free sample Set _ Normal obtained in the step (4.3) to generate a complete machine State sample Set _ State, wherein the complete machine State sample Set _ State meets the Set operation formula:
Set_State＝Set_Fault∪Set_Normal；
(5) establishing and training a fault detection, classification and grading deep neural network fault diagnosis model DNN, which comprises the following specific steps:
(5.1) encoding the sample labels of the samples in the machine State sample Set _ State obtained in the step (4.4) in a binary encoding mode to generate a machine State code, which specifically comprises the following steps:
hierarchical fault type status coding, taking level 5 as an example:
without such failure
Figure BDA0002450121880000071
First stage
Figure BDA0002450121880000072
Second stage
Figure BDA0002450121880000073
Three-stage
Figure BDA0002450121880000074
Four stages
Figure BDA0002450121880000075
Five stages
Figure BDA0002450121880000076
Non-hierarchical fault status encoding: no such fault [0], fault [1 ];
splicing all fault type state codes to obtain a machine state code;
(5.2) determining the structure, the layer number and the node number of the deep neural network fault diagnosis model DNN: the neural network is structurally divided into an input layer of a first layer, a plurality of middle hidden layers and an output layer of a last layer, wherein the input of the input layer is sample characteristics Vec _ t of samples in a machine State sample Set _ State, the sample characteristics Vec _ t are vectors with the dimension of 200-;
(5.3) carrying out unsupervised training on the model established in the step (5.2), and pre-training every two adjacent layers of the DNN as a limited Boltzmann machine to obtain initial model parameters;
(5.4) carrying out supervision training on the initialized DNN by using the initial model parameter obtained by training in the step (5.3), optimizing and micro-adjusting the DNN model parameter by using a back propagation algorithm, wherein the training sample Set is the machine State sample Set _ State obtained in the step (4.4), the sample characteristics Vec _ t of the samples in the Set _ State are input into an input layer, the State codes are output, the training target is to minimize the cross entropy of the output State codes and the State codes generated according to the step (5.1), traversing all samples in the Set _ State, repeating the step, and finally training to obtain a deep neural network fault diagnosis model DNN with unified detection, classification and grading;
(6) and (5) diagnosing the fault of the coal mill to be tested by using the fault detection, classification and grading deep neural network fault diagnosis model DNN obtained in the step (5), and specifically comprising the following steps:
(6.1) generating input sample characteristics of DNN, and specifically comprising the following steps:
(6.1.1) acquiring the rotating speed of the coal mill to be tested in the running process in real time
Figure BDA0002450121880000077
Current of driving motor
Figure BDA0002450121880000078
Voltage of driving motor
Figure BDA0002450121880000079
Coal input
Figure BDA00024501218800000710
Primary air pressure
Figure BDA00024501218800000711
Pressure of pull rod
Figure BDA00024501218800000712
Temperature of
Figure BDA00024501218800000713
Vibration
Figure BDA00024501218800000714
And sound
Figure BDA00024501218800000715
(6.1.2) setting the same data frame length tlen as the step (2), and at the diagnosis time tcurrIntercepting time window tcurr,tcurr+tlen]Inside of
Figure BDA00024501218800000716
And
Figure BDA00024501218800000717
is marked as
Figure BDA00024501218800000718
Figure BDA00024501218800000719
And
Figure BDA00024501218800000721
to obtain tcurrThe data frame at time is recorded as:
Figure BDA0002450121880000081
wherein
Figure BDA0002450121880000082
In the case of a frame of rotational speed data,
Figure BDA0002450121880000083
for driving motor current data frames,
Figure BDA0002450121880000084
A frame of drive motor voltage data,
Figure BDA0002450121880000085
A data frame of coal feeding amount,
Figure BDA0002450121880000086
Is a primary wind pressure data frame,
Figure BDA0002450121880000087
For the pull rod pressure data frame,
Figure BDA0002450121880000088
For the temperature data frame,
Figure BDA0002450121880000089
For the vibration data frame,
Figure BDA00024501218800000810
As frames of sound data
(6.1.3) for the data frame obtained in the step (6.1.2)
Figure BDA00024501218800000811
Processing is carried out to generate sample characteristics, and the specific steps are as follows:
(6.1.3.1) pairs
Figure BDA00024501218800000812
In (1)
Figure BDA00024501218800000813
Figure BDA00024501218800000814
And
Figure BDA00024501218800000815
respectively performing time domain amplitude statistics based on the rotation speed data frame
Figure BDA00024501218800000816
Obtaining the average value of the rotating speed
Figure BDA00024501218800000817
Data frame by drive motor current
Figure BDA00024501218800000818
Obtaining the average value of the current of the driving motor
Figure BDA00024501218800000819
Data frame by driving motor voltage
Figure BDA00024501218800000820
Obtaining the average value of the voltage of the driving motor
Figure BDA00024501218800000821
From the coal-feed data frame
Figure BDA00024501218800000822
Obtaining the average value of coal feeding quantity
Figure BDA00024501218800000823
Derived from a primary wind pressure data frame
Figure BDA00024501218800000824
Mean value of primary air pressure
Figure BDA00024501218800000825
By tension rod pressure data frame
Figure BDA00024501218800000826
Obtaining the average value of the pressure of the pull rod
Figure BDA00024501218800000827
From temperature data frames
Figure BDA00024501218800000828
Obtaining the average value of temperature
Figure BDA00024501218800000829
From frames of vibration data
Figure BDA00024501218800000830
Calculating to obtain the root mean square of vibration
Figure BDA00024501218800000831
Variance of vibration
Figure BDA00024501218800000832
Deviation of vibration
Figure BDA00024501218800000833
And degree of vibration kurtosis
Figure BDA00024501218800000834
From frames of sound data
Figure BDA00024501218800000835
Calculating to obtain sound root mean square
Figure BDA00024501218800000836
Variance of sound
Figure BDA00024501218800000837
Degree of sound skewness
Figure BDA00024501218800000838
And kurtosis of sound
Figure BDA00024501218800000839
The statistics values are spliced into a 15-dimensional vector
Figure BDA00024501218800000840
Is recorded as:
Figure BDA0002450121880000091
(6.1.3.2) pairs
Figure BDA0002450121880000099
In (1)
Figure BDA0002450121880000093
Performing fast Fourier transform to obtain a vibration energy spectrum, performing L subband filtering on the vibration energy spectrum, wherein the value of L is equal to the L value in the step (3.2), and obtaining a L-dimensional vector
Figure BDA0002450121880000094
Is recorded as:
Figure BDA0002450121880000095
wherein FFT represents fast Fourier transform, Filter _ l represents ith subband filtering;
(6.1.3.3) pairs
Figure BDA0002450121880000096
In (1)
Figure BDA0002450121880000097
Performing fast Fourier transform to obtain a sound energy spectrum, then performing M subband filtering on the sound energy spectrum, wherein the value of M is equal to the value of M in the step (3.3), and obtaining an M-dimensional vector
Figure BDA0002450121880000098
Is recorded as:
Figure BDA0002450121880000101
wherein FFT represents fast Fourier transform, Filter _ m represents mth subband filtering;
(6.1.3.4) subjecting the product obtained in the step (6.1.3.1) - (6.1.3.3) to
Figure BDA0002450121880000102
And
Figure BDA0002450121880000103
splicing generates a (15+ L + M) -dimensional vector as tcurrSample characteristics of time of day
Figure BDA0002450121880000104
Is recorded as:
Figure BDA0002450121880000105
(6.2) starting the fault detection, classification and grading deep neural network model DNN obtained by training in the step (5.4), and carrying out analysis on the sample characteristics obtained in the step (6.1.3.4)
Figure BDA0002450121880000106
Input to DNN, output state coding
Figure BDA0002450121880000107
And (3) the value range of each output code is (0, 1), the output codes are judged, if the output value of the output codes is greater than 0.5, the output codes are set to be 1, if the output value of the output codes is less than or equal to 0.5, the output codes are set to be 0, the final diagnosis codes are obtained, the fault-free or fault type and the fault level of the coal mill to be tested corresponding to the final diagnosis codes are obtained according to the state coding rule of the step (5.1), and fault detection, classification and grading of the coal mill based on the neural network unified modeling are achieved.
The invention provides a coal mill fault diagnosis method based on neural network unified modeling, which has the characteristics and advantages that:
compared with the existing coal mill fault diagnosis technology, the diagnosis technology of the invention integrates various signal synthesis running state characteristics of conventional machine production running indexes, machine vibration, machine sound and the like in data use, and the contained data characteristics are richer. The diagnosis method provided by the invention is more intelligent, more accurate in diagnosis and stronger in prediction capability on potential faults. The invention adopts the deep neural network modeling technology to improve the intellectualization of diagnosis and automatically extracts the subtle characteristics which are hidden in the sample data and characterize the fault; the potential fault data with different severity levels are greatly expanded and added on the basis of the basic fault data, so that the fault diagnosis can be predicted in advance. The binary coding is adopted to realize fault detection, classification and grading unified modeling, output nodes are fewer, codes of faults in different grades are arranged in order, and the stability and reliability of diagnosis are improved. In addition, compared with the prior art, the method has a unique advantage that: when multiple faults exist simultaneously, classification and grading unified diagnosis can simultaneously give grading results of various fault types.
Detailed Description
The invention provides a coal mill fault diagnosis method based on neural network unified modeling, which comprises the following steps:
(1) collecting operation state monitoring data of a coal mill to be detected, wherein the operation state monitoring data comprises rotation speed data R, driving motor current I, driving motor voltage E, coal feeding quantity C, primary air pressure W, pull rod pressure P, temperature data T, vibration data V and sound data S of the coal mill to be detected, and R, I, E, C, W, P, T, V and S are time sequence data;
(2) framing the operation state monitoring data acquired in the step (1), setting the duration of a data frame to be tlen, and setting the starting time of the ith data frame to be tiIntercept the time window [ t ]i,ti+tlen]R, I, E, C, W, P, T, V, S in is denoted as R _ ti、I_ti、E_ti、C_ti、W_ti、P_ti、T_ti、V_tiAnd S _ tiAfter the framing processing, R, I, E, C, W, P, T, V, S data are divided into N data frames, which are recorded as:
Figure BDA0002450121880000111
wherein N is the total number of data frames, i is the number of data frames, tiFor the start time of the ith data Frame, Frame _ tiRepresents tiData Frame of time, Frame _ tiFrom tiRotational speed of timeData frame R _ tiAnd a driving motor current data frame I _ tiAnd a driving motor voltage data frame E _ tiCoal feeding amount data frame C _ tiPrimary wind pressure data frame W _ tiPull rod pressure data frame P _ tiTemperature data frame T _ TiVibration data frame V _ tiAnd a sound data frame S _ tiCombining the components;
(3) for the data Frame _ t obtained in the step (2)iProcessing is carried out to generate a sample Feature set Feature _ Full, and the specific steps are as follows:
(3.1) for Frame _ tiR _ t in (1)i、I_ti、E_ti、C_ti、W_ti、P_ti、T_ti、V_tiAnd S _ tiRespectively carrying out time domain amplitude statistics by using a rotating speed data frame R _ tiObtaining the average value R _ mean _ t of the rotating speediFrom the drive motor current data frame I _ tiObtaining the average value I _ mean _ t of the current of the driving motoriVoltage data frame E _ t of driving motoriObtaining the average value E _ mean _ t of the voltage of the driving motoriFrom the coal input data frame C _ tiObtaining the average value C _ mean _ t of the coal feeding quantityiThe primary wind pressure data frame W _ tiObtaining the primary wind pressure average value W _ mean _ tiFrom the pull rod pressure data frame P _ tiObtaining the average value P _ mean _ t of the pressure of the pull rodiFrom the temperature data frame T _ TiObtaining the temperature average value T _ mean _ TiFrom the vibration data frame V _ tiCalculating to obtain a vibration root mean square V _ rms _ tiVibration variance V _ sigm _ tiVibration skewness V _ skew _ tiAnd the vibration kurtosis V _ kurt _ tiFrom the sound data frame S _ tiCalculating to obtain the root mean square S _ rms _ t of the soundiSound variance S _ sigm _ tiSound skewness S _ skew _ tiAnd sound kurtosis S _ kurt _ tiThe statistics values are spliced into a 15-dimensional vector Vec1_ tiN Frame _ tiThe data frame processing obtains N15-dimensional vectors, which are recorded as:
Figure BDA0002450121880000121
(3.2) for Frame _ tiV _ t in (1)iPerforming fast Fourier transform to obtain a vibration energy spectrum, and performing L subband filtering on the vibration energy spectrum to obtain a L-dimensional vector Vec2_ tiN number of V _ tiThe data frame processing yields N L-dimensional vectors, denoted as:
Figure BDA0002450121880000131
wherein L is the number of Filter subbands, L is a value range of 10-1000, FFT represents fast fourier transform, Filter _ l represents the first subband filtering;
(3.3) for Frame _ tiS _ t in (1)iPerforming fast Fourier transform to obtain a sound energy spectrum, and performing M-subband filtering on the sound energy spectrum to obtain an M-dimensional vector Vec3_ tiN S _ tiProcessing the data frame to obtain N M-dimensional vectors, which are recorded as:
Figure BDA0002450121880000132
wherein M is the number of filtering sub-bands, the value of M is 10-1000, FFT represents fast Fourier transform, Filter _ M represents mth sub-band filtering;
(3.4) mixing Vec1_ t obtained in the steps (3.1) to (3.3)i、Vec2_tiAnd Vec3_ tiSplicing generates a (15+ L + M) -dimensional vector as tiSample characteristic Vec _ t of time instantiAnd is recorded as:
Figure BDA0002450121880000133
(3.5) obtaining N sample characteristics Vec _ t obtained in the step (3.4)iThe set constitutes a sample Feature set Feature _ Full, noted as:
Figure BDA0002450121880000134
(4) training sample set generation, comprising the steps of:
(4.1) dividing the machine status into a fault status and a no fault status, wherein the fault status comprises a plurality of non-classified fault types and a plurality of classified fault type statuses, and dividing the classified fault types into 5 levels according to severity degrees: wherein level 1 indicates in early stage of the fault, level 2 indicates in early stage of the fault, level 3 indicates in middle stage of the fault, level 4 indicates in middle and late stage of the fault, level 5 indicates in late stage of the fault, and the fault duration length vectors of 5 fault levels are represented as [ D1, D2, D3, D4, D5 ]; wherein D1 represents the duration length of progression from level 1 to level 2, D2 represents the duration length of progression from level 2 to level 3, D3 represents the duration length of progression from level 3 to level 4, D4 represents the duration length of progression from level 4 to level 5, D5 represents the duration length from discovery of D5 signature to occurrence of a destructive fault, each classified fault type determines the fault duration length vector as [ D1_ type, D2_ type, D3_ type, D4_ type, D5_ type ], which represents the fault type number in the manner described above;
(4.2) generating a Fault sample Set _ Fault according to a machine Fault marking log obtained from a machine operation maintenance management department from the sample Feature Set Feature _ Full obtained in the step (3.5), wherein the method specifically comprises the following steps:
(4.2.1) extracting a record from the machine fault labeling log, wherein the record content comprises a group of quaternary data in the form of: (τ, type, level, τ)2) Wherein tau is the time when the fault is detected, type is the fault type, level is the fault level, tau is2The moment when the fault is repaired;
(4.2.2) judging the fault type:
if type is not classified fault type, executing steps (4.2.2.1) - (4.2.2.2):
(4.2.2.1) extracting all sample features Vec _ t in a period of time when tau is less than or equal to t < tau 2 from the sample Feature set Feature _ Full obtained in the step (3.5);
(4.2.2.2) adding a Sample label (type) to each Sample feature Vec _ t obtained in the step (4.2.2.1) to generate a labeled fault Sample _ fault ═ (Vec _ t, (type)), wherein Sample _ fault represents a labeled fault Sample, Vec _ t is a Sample feature, (type) is a Sample label, type is a fault type, and "×" is a default item;
if the type is a classified fault type, setting the number of the fault type as itype, and executing the steps (4.2.2.3) - (4.2.2.4):
(4.2.2.3) τ, level, τ obtained according to step (4.2.1)2And the fault duration length vector [ D1_ type, D2_ type, D3_ type, D4_ type, D5_ type ] obtained in the step (4.1)]When the conversion time of the faults in different levels is calculated, taking level as 5 as an example, and level is 4, 3, 2, 1, the processing method is analogized, and the result is as follows:
Figure BDA0002450121880000151
(4.2.2.4) processing each Sample vec _ t in the Sample Feature set Feature _ Full obtained in step (3.5) as follows to generate a labeled fault Sample _ fault:
Figure BDA0002450121880000152
(4.2.3) traversing each record in the machine Fault labeling log, circularly executing the steps (4.2.1) and (4.2.2), and combining all marked Fault samples generated by the steps (4.2.2.2) and (4.2.2.4) into a Set to obtain a Fault sample Set _ Fault;
(4.3) generating a fault-free sample Set _ Normal, comprising the steps of:
(4.3.1) marking the sample Feature corresponding to the Fault sample Set _ Fault generated in the step (4.2.3) as a Fault sample Feature _ Fault, removing the Feature _ Fault from the sample Feature Set Feature _ Full obtained in the step (3.5), marking the residual sample Feature Set as a non-Fault sample Feature Set Feature _ Normal, and meeting a Set operation formula:
Feature_Normal＝Feature_Full-Feature_fault
(4.3.2) randomly extracting a sample Feature vec _ t from the failure-free sample Feature set Feature _ Normal obtained in the step (4.3.1), adding a sample label, and generating a labeled failure-free sample record:
sample _ normal ═ (vec _ t, (no fault type,));
wherein, Sample _ normal is a labeled no-fault Sample, vec _ t is a Sample characteristic, (no-fault type, is a Sample label), "no-fault type" is a machine state type, and "+" is a default term;
(4.3.3) circularly executing the step (4.3.2) to obtain marked Fault-free samples with the number equal to that of the samples of the Fault sample Set _ Fault, and collecting to form a Fault-free sample Set _ Normal;
(4.4) merging the Fault sample Set _ Fault obtained in the step (4.2) with the Fault-free sample Set _ Normal obtained in the step (4.3) to generate a complete machine State sample Set _ State, wherein the complete machine State sample Set _ State meets the Set operation formula:
Set_State＝Set_Fault∪Set_Normal；
(5) establishing and training a fault detection, classification and grading deep neural network fault diagnosis model DNN, which comprises the following specific steps:
(5.1) encoding the sample labels of the samples in the machine State sample Set _ State obtained in the step (4.4) in a binary encoding mode to generate a machine State code, which specifically comprises the following steps:
hierarchical fault type status coding, taking level 5 as an example:
without such failure
Figure BDA0002450121880000161
First stage
Figure BDA0002450121880000162
Second stage
Figure BDA0002450121880000163
Three-stage
Figure BDA0002450121880000164
Four stages
Figure BDA0002450121880000165
Five stages
Figure BDA0002450121880000166
Non-hierarchical fault status encoding: no such fault [0], fault [1 ];
splicing all fault type state codes to obtain a machine state code;
taking 2 classified fault types (type 1 and type 2) and 1 unclassified type (type 3) as examples, the machine state coding form after splicing is as follows:
[c1 c2 c3 c4 c5 c6 c7]
wherein [ c1 c2 c3] is type 1 state code, [ c4 c5 c6] is type 2 state code, [ c7] is type 3 state code. The machine state (fault, fault type and level thereof) can be determined according to the state coding form of [ c1 c2 c3 c4 c5 c6 c7], wherein [ 0000000 ] represents the fault-free state without any fault type; if multiple fault types coexist, the state code can indicate multiple fault types simultaneously;
(5.2) determining the structure, the layer number and the node number of the deep neural network fault diagnosis model DNN: the neural network is structurally divided into an input layer of a first layer, a plurality of middle hidden layers and an output layer of a last layer, wherein the input of the input layer is a sample characteristic Vec _ t of a sample in a machine State sample Set _ State, the sample characteristic Vec _ t is a vector with the dimension of 200-:
Node1＝1024
Node2＝512
Node3＝256
Node4＝512
Node5＝1024
the network structure adopts a structural form that no connection exists in layers and adjacent layers are fully connected;
(5.3) modeling of (5.2)Carrying out unsupervised training, and pre-training every two adjacent layers of DNN as a limited Boltzmann machine to obtain initial model parameters; taking a 7-layer neural network with 5 hidden layers as an example, 6 limited boltzmann machines (RBMs) need to be connected in total: firstly, training RBM1 composed of layer 1 and layer 2 to obtain model parameters w12And b2(ii) a Then training to obtain RBM2 composed of layer 2 and layer 3, and obtaining model parameter w23And b3(ii) a Sequentially executing to obtain parameters of all 6 limited Boltzmann machines, and pre-training the obtained parameters to form initial model parameters (w) of DNN (Dempster-Namex) by the 6 limited Boltzmann machines12,b2,w23,b3,w34,b4,w45,b5,w56,b6,w67,b7)；
(5.4) carrying out supervision training on the initialized DNN by using the initial model parameter obtained by training in the step (5.3), optimizing and micro-adjusting the DNN model parameter by using a back propagation algorithm, wherein the training sample Set is the machine State sample Set _ State obtained in the step (4.4), the sample characteristics Vec _ t of the samples in the Set _ State are input into an input layer, the State codes are output, the training target is to minimize the cross entropy of the output State codes and the State codes generated according to the step (5.1), traversing all samples in the Set _ State, repeating the step, and finally training to obtain a deep neural network fault diagnosis model DNN with unified detection, classification and grading;
(6) and (5) diagnosing the fault of the coal mill to be tested by using the fault detection, classification and grading deep neural network fault diagnosis model DNN obtained in the step (5), and specifically comprising the following steps:
(6.1) generating input sample characteristics of DNN, and specifically comprising the following steps:
(6.1.1) acquiring the rotating speed of the coal mill to be tested in the running process in real time
Figure BDA0002450121880000171
Current of driving motor
Figure BDA0002450121880000172
Voltage of driving motor
Figure BDA0002450121880000173
Coal input
Figure BDA0002450121880000174
Primary air pressure
Figure BDA0002450121880000175
Pressure of pull rod
Figure BDA0002450121880000176
Temperature of
Figure BDA0002450121880000177
Vibration
Figure BDA0002450121880000178
And sound
Figure BDA0002450121880000179
(6.1.2) setting the same data frame length tlen as the step (2), and at the diagnosis time tcurrIntercepting time window tcurr,tcurr+tlen]Inside of
Figure BDA00024501218800001710
And
Figure BDA00024501218800001711
is marked as
Figure BDA00024501218800001712
Figure BDA00024501218800001713
And
Figure BDA00024501218800001714
to obtain tcurrThe data frame at time is recorded as:
Figure BDA0002450121880000181
wherein
Figure BDA0002450121880000182
In the case of a frame of rotational speed data,
Figure BDA0002450121880000183
for driving motor current data frames,
Figure BDA0002450121880000184
A frame of drive motor voltage data,
Figure BDA0002450121880000185
A data frame of coal feeding amount,
Figure BDA0002450121880000186
Is a primary wind pressure data frame,
Figure BDA0002450121880000187
For the pull rod pressure data frame,
Figure BDA0002450121880000188
For the temperature data frame,
Figure BDA0002450121880000189
For the vibration data frame,
Figure BDA00024501218800001841
As frames of sound data
(6.1.3) for the data frame obtained in the step (6.1.2)
Figure BDA00024501218800001811
Processing is carried out to generate sample characteristics, and the specific steps are as follows:
(6.1.3.1) pairs
Figure BDA00024501218800001812
In (1)
Figure BDA00024501218800001813
Figure BDA00024501218800001814
And
Figure BDA00024501218800001815
respectively performing time domain amplitude statistics based on the rotation speed data frame
Figure BDA00024501218800001816
Obtaining the average value of the rotating speed
Figure BDA00024501218800001817
Data frame by drive motor current
Figure BDA00024501218800001818
Obtaining the average value of the current of the driving motor
Figure BDA00024501218800001819
Data frame by driving motor voltage
Figure BDA00024501218800001820
Obtaining the average value of the voltage of the driving motor
Figure BDA00024501218800001821
From the coal-feed data frame
Figure BDA00024501218800001822
Obtaining the average value of coal feeding quantity
Figure BDA00024501218800001823
Derived from a primary wind pressure data frame
Figure BDA00024501218800001824
Mean value of primary air pressure
Figure BDA00024501218800001825
By tension rod pressure data frame
Figure BDA00024501218800001826
Obtaining the average value of the pressure of the pull rod
Figure BDA00024501218800001827
From temperature data frames
Figure BDA00024501218800001828
Obtaining the average value of temperature
Figure BDA00024501218800001829
From frames of vibration data
Figure BDA00024501218800001830
Calculating to obtain the root mean square of vibration
Figure BDA00024501218800001831
Variance of vibration
Figure BDA00024501218800001832
Deviation of vibration
Figure BDA00024501218800001833
And degree of vibration kurtosis
Figure BDA00024501218800001834
From frames of sound data
Figure BDA00024501218800001835
Calculating to obtain sound root mean square
Figure BDA00024501218800001836
Variance of sound
Figure BDA00024501218800001837
Degree of sound skewness
Figure BDA00024501218800001838
And kurtosis of sound
Figure BDA00024501218800001839
The statistics values are spliced into a 15-dimensional vector
Figure BDA00024501218800001840
Is recorded as:
Figure BDA0002450121880000191
(6.1.3.2) pairs
Figure BDA0002450121880000192
In (1)
Figure BDA0002450121880000193
Performing fast Fourier transform to obtain a vibration energy spectrum, performing L subband filtering on the vibration energy spectrum, wherein the value of L is equal to the L value in the step (3.2), and obtaining a L-dimensional vector
Figure BDA0002450121880000194
Is recorded as:
Figure BDA0002450121880000195
wherein FFT represents fast Fourier transform, Filter _ l represents ith subband filtering;
(6.1.3.3) pairs
Figure BDA0002450121880000196
In (1)
Figure BDA0002450121880000197
Performing fast Fourier transform to obtain a sound energy spectrum, then performing M subband filtering on the sound energy spectrum, wherein the value of M is equal to the value of M in the step (3.3), and obtaining an M-dimensional vector
Figure BDA0002450121880000198
Is recorded as:
Figure BDA0002450121880000201
wherein FFT represents fast Fourier transform, Filter _ m represents mth subband filtering;
(6.1.3.4) subjecting the product obtained in the step (6.1.3.1) - (6.1.3.3) to
Figure BDA0002450121880000202
And
Figure BDA0002450121880000203
splicing generates a (15+ L + M) -dimensional vector as tcurrSample characteristics of time of day
Figure BDA0002450121880000204
Is recorded as:
Figure BDA0002450121880000205
(6.2) starting the fault detection, classification and grading deep neural network model DNN obtained by training in the step (5.4), and carrying out analysis on the sample characteristics obtained in the step (6.1.3.4)
Figure BDA0002450121880000206
Input to DNN, output state coding
Figure BDA0002450121880000207
And (3) the value range of each output code is (0, 1), the output codes are judged, if the output value of the output codes is greater than 0.5, the output codes are set to be 1, if the output value of the output codes is less than or equal to 0.5, the output codes are set to be 0, the final diagnosis codes are obtained, the fault-free or fault type and the fault level of the coal mill to be tested corresponding to the final diagnosis codes are obtained according to the state coding rule of the step (5.1), and fault detection, classification and grading of the coal mill based on the neural network unified modeling are achieved.

Claims (1)

1. A coal mill fault diagnosis method based on neural network unified modeling is characterized by comprising the following steps:
(1) collecting operation state monitoring data of a coal mill to be detected, wherein the operation state monitoring data comprises rotation speed data R, driving motor current I, driving motor voltage E, coal feeding quantity C, primary air pressure W, pull rod pressure P, temperature data T, vibration data V and sound data S of the coal mill to be detected, and R, I, E, C, W, P, T, V and S are time sequence data;
(2) framing the operation state monitoring data acquired in the step (1), setting the duration of a data frame to be tlen, and setting the starting time of the ith data frame to be tiIntercept the time window [ t ]i,ti+tlen]R, I, E, C, W, P, T, V, S in is denoted as R _ ti、I_ti、E_ti、C_ti、W_ti、P_ti、T_ti、V_tiAnd S _ tiAfter the framing processing, R, I, E, C, W, P, T, V, S data are divided into N data frames, which are recorded as:
Figure FDA0002450121870000011
wherein N is the total number of data frames, i is the number of data frames, tiFor the start time of the ith data Frame, Frame _ tiRepresents tiData Frame of time, Frame _ tiFrom tiFrame of rotational speed data at time R _ tiAnd a driving motor current data frame I _ tiAnd a driving motor voltage data frame E _ tiCoal feeding amount data frame C _ tiPrimary wind pressure data frame W _ tiPull rod pressure data frame P _ tiTemperature data frame T _ TiVibration data frame V _ tiAnd a sound data frame S _ tiCombining the components;
(3) for the data Frame _ t obtained in the step (2)iProcessing is carried out to generate a sample Feature set Feature _ Full, and the specific steps are as follows:
(3.1) for Frame _ tiR _ t in (1)i、I_ti、E_ti、C_ti、W_ti、P_ti、T_ti、V_tiAnd S _ tiRespectively carrying out time domain amplitude statistics by using a rotating speed data frame R _ tiObtaining the average value R _ mean _ t of the rotating speediFrom the drive motor current data frame I _ tiObtaining the average value I _ mean _ t of the current of the driving motoriVoltage data frame E _ t of driving motoriObtaining the average value E _ m of the voltage of the driving motorean_tiFrom the coal input data frame C _ tiObtaining the average value C _ mean _ t of the coal feeding quantityiThe primary wind pressure data frame W _ tiObtaining the primary wind pressure average value W _ mean _ tiFrom the pull rod pressure data frame P _ tiObtaining the average value P _ mean _ t of the pressure of the pull rodiFrom the temperature data frame T _ TiObtaining the temperature average value T _ mean _ TiFrom the vibration data frame V _ tiCalculating to obtain a vibration root mean square V _ rms _ tiVibration variance V _ sigm _ tiVibration skewness V _ skew _ tiAnd the vibration kurtosis V _ kurt _ tiFrom the sound data frame S _ tiCalculating to obtain the root mean square S _ rms _ t of the soundiSound variance S _ sigm _ tiSound skewness S _ skew _ tiAnd sound kurtosis S _ kurt _ tiThe statistics values are spliced into a 15-dimensional vector Vec1_ tiN Frame _ tiThe data frame processing obtains N15-dimensional vectors, which are recorded as:
Figure FDA0002450121870000021
(3.2) for Frame _ tiV _ t in (1)iPerforming fast Fourier transform to obtain a vibration energy spectrum, and performing L subband filtering on the vibration energy spectrum to obtain a L-dimensional vector Vec2_ tiN number of V _ tiThe data frame processing yields N L-dimensional vectors, denoted as:
Figure FDA0002450121870000022
wherein L is the number of Filter subbands, L is a value range of 10-1000, FFT represents fast fourier transform, Filter _ l represents the first subband filtering;
(3.3) for Frame _ tiS _ t in (1)iPerforming fast Fourier transform to obtain a sound energy spectrum, and performing M-subband filtering on the sound energy spectrum to obtain an M-dimensional vector Vec3_ tiN S _ tiProcessing the data frame to obtain N M-dimensional vectors, which are recorded as:
Figure FDA0002450121870000031
wherein M is the number of filtering sub-bands, the value of M is 10-1000, FFT represents fast Fourier transform, Filter _ M represents mth sub-band filtering;
(3.4) mixing Vec1_ t obtained in the steps (3.1) to (3.3)i、Vec2_tiAnd Vec3_ tiSplicing generates a (15+ L + M) -dimensional vector as tiSample characteristic Vec _ t of time instantiAnd is recorded as:
Figure FDA0002450121870000032
(3.5) obtaining N sample characteristics Vec _ t obtained in the step (3.4)iThe set constitutes a sample Feature set Feature _ Full, noted as:
Figure FDA0002450121870000033
(4) training sample set generation, comprising the steps of:
(4.1) dividing the machine status into a fault status and a no fault status, wherein the fault status comprises a plurality of non-classified fault types and a plurality of classified fault type statuses, and dividing the classified fault types into 5 levels according to severity degrees: wherein level 1 indicates in early stage of the fault, level 2 indicates in early stage of the fault, level 3 indicates in middle stage of the fault, level 4 indicates in middle and late stage of the fault, level 5 indicates in late stage of the fault, and the fault duration length vectors of 5 fault levels are represented as [ D1, D2, D3, D4, D5 ]; wherein D1 represents the duration length of progression from level 1 to level 2, D2 represents the duration length of progression from level 2 to level 3, D3 represents the duration length of progression from level 3 to level 4, D4 represents the duration length of progression from level 4 to level 5, D5 represents the duration length from discovery of D5 signature to occurrence of a destructive fault, each classified fault type determines the fault duration length vector as [ D1_ type, D2_ type, D3_ type, D4_ type, D5_ type ], which represents the fault type number in the manner described above;
(4.2) generating a Fault sample Set _ Fault according to a machine Fault marking log obtained from a machine operation maintenance management department from the sample Feature Set Feature _ Full obtained in the step (3.5), wherein the method specifically comprises the following steps:
(4.2.1) extracting a record from the machine fault labeling log, wherein the record content comprises a group of quaternary data in the form of: (τ, type, level, τ)2) Wherein tau is the time when the fault is detected, type is the fault type, level is the fault level, tau is2The moment when the fault is repaired;
(4.2.2) judging the fault type:
if type is not classified fault type, executing steps (4.2.2.1) - (4.2.2.2):
(4.2.2.1) extracting all sample features Vec _ t in a period of time when tau is less than or equal to t < tau 2 from the sample Feature set Feature _ Full obtained in the step (3.5);
(4.2.2.2) adding a Sample label (type) to each Sample feature Vec _ t obtained in the step (4.2.2.1) to generate a labeled fault Sample _ fault ═ (Vec _ t, (type)), wherein Sample _ fault represents a labeled fault Sample, Vec _ t is a Sample feature, (type) is a Sample label, type is a fault type, and "×" is a default item;
if the type is a classified fault type, setting the number of the fault type as itype, and executing the steps (4.2.2.3) - (4.2.2.4):
(4.2.2.3) τ, level, τ obtained according to step (4.2.1)2And the fault duration length vector [ D1_ type, D2_ type, D3_ type, D4_ type, D5_ type ] obtained in the step (4.1)]When the conversion time of the faults in different levels is calculated, taking level as 5 as an example, and level is 4, 3, 2, 1, the processing method is analogized, and the result is as follows:
Figure FDA0002450121870000041
(4.2.2.4) processing each Sample vec _ t in the Sample Feature set Feature _ Full obtained in step (3.5) as follows to generate a labeled fault Sample _ fault:
Figure FDA0002450121870000051
(4.2.3) traversing each record in the machine Fault labeling log, circularly executing the steps (4.2.1) and (4.2.2), and combining all marked Fault samples generated by the steps (4.2.2.2) and (4.2.2.4) into a Set to obtain a Fault sample Set _ Fault;
(4.3) generating a fault-free sample Set _ Normal, comprising the steps of:
(4.3.1) marking the sample Feature corresponding to the Fault sample Set _ Fault generated in the step (4.2.3) as a Fault sample Feature _ Fault, removing the Feature _ Fault from the sample Feature Set Feature _ Full obtained in the step (3.5), marking the residual sample Feature Set as a non-Fault sample Feature Set Feature _ Normal, and meeting a Set operation formula:
Feature_Normal＝Feature_Full-Feature_fault
(4.3.2) randomly extracting a sample Feature vec _ t from the failure-free sample Feature set Feature _ Normal obtained in the step (4.3.1), adding a sample label, and generating a labeled failure-free sample record:
sample _ normal ═ (vec _ t, (no fault type,));
wherein, Sample _ normal is a labeled no-fault Sample, vec _ t is a Sample characteristic, (no-fault type, is a Sample label), "no-fault type" is a machine state type, and "+" is a default term;
(4.3.3) circularly executing the step (4.3.2) to obtain marked Fault-free samples with the number equal to that of the samples of the Fault sample Set _ Fault, and collecting to form a Fault-free sample Set _ Normal;
(4.4) merging the Fault sample Set _ Fault obtained in the step (4.2) with the Fault-free sample Set _ Normal obtained in the step (4.3) to generate a complete machine State sample Set _ State, wherein the complete machine State sample Set _ State meets the Set operation formula:
Set_State＝Set_Fault∪Set_Normal；
(5) establishing and training a fault detection, classification and grading deep neural network fault diagnosis model DNN, which comprises the following specific steps:
(5.1) encoding the sample labels of the samples in the machine State sample Set _ State obtained in the step (4.4) in a binary encoding mode to generate a machine State code, which specifically comprises the following steps:
hierarchical fault type status coding, taking level 5 as an example:
without such failure
Figure FDA0002450121870000061
First stage
Figure FDA0002450121870000062
Second stage
Figure FDA0002450121870000063
Three-stage
Figure FDA0002450121870000064
Four stages
Figure FDA0002450121870000065
Five stages
Figure FDA0002450121870000066
Non-hierarchical fault status encoding: no such fault [0], fault [1 ];
splicing all fault type state codes to obtain a machine state code;
(5.2) determining the structure, the layer number and the node number of the deep neural network fault diagnosis model DNN: the neural network is structurally divided into an input layer of a first layer, a plurality of middle hidden layers and an output layer of a last layer, wherein the input of the input layer is sample characteristics Vec _ t of samples in a machine State sample Set _ State, the sample characteristics Vec _ t are vectors with the dimension of 200-;
(5.3) carrying out unsupervised training on the model established in the step (5.2), and pre-training every two adjacent layers of the DNN as a limited Boltzmann machine to obtain initial model parameters;
(5.4) carrying out supervision training on the initialized DNN by using the initial model parameter obtained by training in the step (5.3), optimizing and micro-adjusting the DNN model parameter by using a back propagation algorithm, wherein the training sample Set is the machine State sample Set _ State obtained in the step (4.4), the sample characteristics Vec _ t of the samples in the Set _ State are input into an input layer, the State codes are output, the training target is to minimize the cross entropy of the output State codes and the State codes generated according to the step (5.1), traversing all samples in the Set _ State, repeating the step, and finally training to obtain a deep neural network fault diagnosis model DNN with unified detection, classification and grading;
(6) and (5) diagnosing the fault of the coal mill to be tested by using the fault detection, classification and grading deep neural network fault diagnosis model DNN obtained in the step (5), and specifically comprising the following steps:
(6.1) generating input sample characteristics of DNN, and specifically comprising the following steps:
(6.1.1) acquiring the rotating speed of the coal mill to be tested in the running process in real time
Figure FDA0002450121870000067
Current of driving motor
Figure FDA0002450121870000068
Voltage of driving motor
Figure FDA0002450121870000069
Coal input
Figure FDA00024501218700000610
Primary air pressure
Figure FDA00024501218700000611
Pressure of pull rod
Figure FDA00024501218700000612
Temperature of
Figure FDA00024501218700000613
Vibration
Figure FDA00024501218700000614
And sound
Figure FDA00024501218700000615
(6.1.2) setting the same data frame length tlen as the step (2), and at the diagnosis time tcurrIntercepting time window tcurr,tcurr+tlen]Inside of
Figure FDA0002450121870000071
And
Figure FDA0002450121870000072
is marked as
Figure FDA0002450121870000073
Figure FDA0002450121870000074
And
Figure FDA0002450121870000075
to obtain tcurrThe data frame at time is recorded as:
Figure FDA0002450121870000076
wherein
Figure FDA0002450121870000077
In the case of a frame of rotational speed data,
Figure FDA0002450121870000078
for driving motor current data frame、
Figure FDA0002450121870000079
A frame of drive motor voltage data,
Figure FDA00024501218700000710
A data frame of coal feeding amount,
Figure FDA00024501218700000711
Is a primary wind pressure data frame,
Figure FDA00024501218700000712
For the pull rod pressure data frame,
Figure FDA00024501218700000713
For the temperature data frame,
Figure FDA00024501218700000714
For the vibration data frame,
Figure FDA00024501218700000715
As frames of sound data
(6.1.3) for the data frame obtained in the step (6.1.2)
Figure FDA00024501218700000716
Processing is carried out to generate sample characteristics, and the specific steps are as follows:
(6.1.3.1) pairs
Figure FDA00024501218700000717
In (1)
Figure FDA00024501218700000718
Figure FDA00024501218700000719
And
Figure FDA00024501218700000720
respectively performing time domain amplitude statistics based on the rotation speed data frame
Figure FDA00024501218700000721
Obtaining the average value of the rotating speed
Figure FDA00024501218700000722
Data frame by drive motor current
Figure FDA00024501218700000723
Obtaining the average value of the current of the driving motor
Figure FDA00024501218700000724
Data frame by driving motor voltage
Figure FDA00024501218700000725
Obtaining the average value of the voltage of the driving motor
Figure FDA00024501218700000726
From the coal-feed data frame
Figure FDA00024501218700000727
Obtaining the average value of coal feeding quantity
Figure FDA00024501218700000728
Derived from a primary wind pressure data frame
Figure FDA00024501218700000729
Mean value of primary air pressure
Figure FDA00024501218700000730
By tension rod pressure data frame
Figure FDA00024501218700000731
Obtaining the average value of the pressure of the pull rod
Figure FDA00024501218700000732
From temperature data frames
Figure FDA00024501218700000733
Obtaining the average value of temperature
Figure FDA00024501218700000734
From frames of vibration data
Figure FDA00024501218700000735
Calculating to obtain the root mean square of vibration
Figure FDA00024501218700000736
Variance of vibration
Figure FDA00024501218700000737
Deviation of vibration
Figure FDA00024501218700000738
And degree of vibration kurtosis
Figure FDA00024501218700000739
From frames of sound data
Figure FDA00024501218700000740
Calculating to obtain sound root mean square
Figure FDA00024501218700000741
Variance of sound
Figure FDA00024501218700000742
Degree of sound skewness
Figure FDA00024501218700000743
And kurtosis of sound
Figure FDA0002450121870000081
Will be the above-mentioned systemThe evaluation value is spliced to form a 15-dimensional vector
Figure FDA0002450121870000082
Is recorded as:
Figure FDA0002450121870000083
(6.1.3.2) pairs
Figure FDA0002450121870000084
In (1)
Figure FDA0002450121870000085
Performing fast Fourier transform to obtain a vibration energy spectrum, performing L subband filtering on the vibration energy spectrum, wherein the value of L is equal to the L value in the step (3.2), and obtaining a L-dimensional vector
Figure FDA0002450121870000086
Is recorded as:
Figure FDA0002450121870000087
wherein FFT represents fast Fourier transform, Filter _ l represents ith subband filtering;
(6.1.3.3) pairs
Figure FDA0002450121870000088
In (1)
Figure FDA0002450121870000089
Performing fast Fourier transform to obtain a sound energy spectrum, then performing M subband filtering on the sound energy spectrum, wherein the value of M is equal to the value of M in the step (3.3), and obtaining an M-dimensional vector
Figure FDA00024501218700000810
Is recorded as:
Figure FDA0002450121870000091
wherein FFT represents fast Fourier transform, Filter _ m represents mth subband filtering;
(6.1.3.4) subjecting the product obtained in the step (6.1.3.1) - (6.1.3.3) to
Figure FDA0002450121870000092
And
Figure FDA0002450121870000093
splicing generates a (15+ L + M) -dimensional vector as tcurrSample characteristics of time of day
Figure FDA0002450121870000094
Is recorded as:
Figure FDA0002450121870000095
(6.2) starting the fault detection, classification and grading deep neural network model DNN obtained by training in the step (5.4), and carrying out analysis on the sample characteristics obtained in the step (6.1.3.4)
Figure FDA0002450121870000096
Input to DNN, output state coding
Figure FDA0002450121870000097
And (3) the value range of each output code is (0, 1), the output codes are judged, if the output value of the output codes is greater than 0.5, the output codes are set to be 1, if the output value of the output codes is less than or equal to 0.5, the output codes are set to be 0, the final diagnosis codes are obtained, the fault-free or fault type and the fault level of the coal mill to be tested corresponding to the final diagnosis codes are obtained according to the state coding rule of the step (5.1), and fault detection, classification and grading of the coal mill based on the neural network unified modeling are achieved.
CN202010290250.9A 2020-04-14 2020-04-14 Coal mill fault diagnosis method based on neural network unified modeling Withdrawn CN111488935A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010290250.9A CN111488935A (en) 2020-04-14 2020-04-14 Coal mill fault diagnosis method based on neural network unified modeling

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010290250.9A CN111488935A (en) 2020-04-14 2020-04-14 Coal mill fault diagnosis method based on neural network unified modeling

Publications (1)

Publication Number Publication Date
CN111488935A true CN111488935A (en) 2020-08-04

Family

ID=71798568

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010290250.9A Withdrawn CN111488935A (en) 2020-04-14 2020-04-14 Coal mill fault diagnosis method based on neural network unified modeling

Country Status (1)

Country Link
CN (1) CN111488935A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112800563A (en) * 2021-03-30 2021-05-14 三一重型装备有限公司 Coal mining machine fault discrimination method and system and readable storage medium
CN113553465A (en) * 2021-06-15 2021-10-26 深圳供电局有限公司 Sound data storage method and device, computer equipment and storage medium
CN114021855A (en) * 2021-12-02 2022-02-08 清华大学 Historical data-based traction motor temperature rise prediction method
CN114492938A (en) * 2021-12-29 2022-05-13 中国大唐集团科学技术研究总院有限公司华东电力试验研究院 Coal mill fault early warning method and system based on BPNN model and adaptive threshold
CN117033981A (en) * 2023-10-07 2023-11-10 广东电网有限责任公司阳江供电局 Submarine cable fault detection method, submarine cable fault detection device, submarine cable fault detection equipment and submarine cable fault detection medium

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112800563A (en) * 2021-03-30 2021-05-14 三一重型装备有限公司 Coal mining machine fault discrimination method and system and readable storage medium
CN113553465A (en) * 2021-06-15 2021-10-26 深圳供电局有限公司 Sound data storage method and device, computer equipment and storage medium
CN113553465B (en) * 2021-06-15 2023-12-19 深圳供电局有限公司 Sound data warehousing method, device, computer equipment and storage medium
CN114021855A (en) * 2021-12-02 2022-02-08 清华大学 Historical data-based traction motor temperature rise prediction method
CN114492938A (en) * 2021-12-29 2022-05-13 中国大唐集团科学技术研究总院有限公司华东电力试验研究院 Coal mill fault early warning method and system based on BPNN model and adaptive threshold
CN117033981A (en) * 2023-10-07 2023-11-10 广东电网有限责任公司阳江供电局 Submarine cable fault detection method, submarine cable fault detection device, submarine cable fault detection equipment and submarine cable fault detection medium
CN117033981B (en) * 2023-10-07 2024-02-09 广东电网有限责任公司阳江供电局 Submarine cable fault detection method, submarine cable fault detection device, submarine cable fault detection equipment and submarine cable fault detection medium

Similar Documents

Publication Publication Date Title
CN111488935A (en) Coal mill fault diagnosis method based on neural network unified modeling
CN111428685A (en) Machine fault detection, classification and grading method based on neural network unified modeling
Zhang et al. Fault diagnosis and prognosis using wavelet packet decomposition, Fourier transform and artificial neural network
CN111523659A (en) Machine fault prediction diagnosis method based on three-level neural network modeling
CN111596604A (en) Intelligent fault diagnosis and self-healing control system and method for engineering equipment based on digital twinning
CN111722145B (en) Synchronous motor excitation winding turn-to-turn short circuit mild fault diagnosis method
CN110766175A (en) Pitch system fault detection method and device based on optimal interval distribution machine
CN110040594A (en) A kind of elevator operation detection system and method based on convolutional neural networks
CN116383739B (en) Intelligent fault diagnosis method based on domain self-adaption multi-mode data fusion
CN111795819A (en) Gear box fault diagnosis method integrating vibration and current signal collaborative learning
Shim et al. Unsupervised anomaly detection for electric drives based on variational auto-encoder
Zhang et al. A gan based fault detection of wind turbines gearbox
CN114818779B (en) Fan rotor fault diagnosis method based on hierarchical classification algorithm
CN114997749B (en) Intelligent scheduling method and system for power personnel
Yang et al. Anomaly detection of automobile welding workstation based on stacked autoencoders and long short-term memory network
CN115560962A (en) Fan abnormal state detection method based on multi-level space-time diagram neural network
Baptista et al. One-Class Classifiers for Novelties Detection in Electrical Submersible Pumps
Zhang et al. A Fault Diagnosis Method for Electrical Equipment With Imbalanced SCADA Data Based on SMOTE Oversampling and Domain Adaptation
Sobha et al. A comprehensive approach for gearbox fault detection and diagnosis using sequential neural networks
CN118606873B (en) Two-stage anomaly classification method based on self-encoder
Duhaney et al. Studying the effect of class imbalance in ocean turbine fault data on reliable state detection
Yu et al. A Hybrid Intelligent Method for Rolling Bearing Fault Diagnosis Integrated with Expert Knowledge and Deep Learning
Yu et al. Research on AR modeling method with SOFM-based classifier applied to gear multi-faults diagnosis
Han et al. Diagnosis of motor aging through CNN model using signal correlation
Zhuang et al. Deep Learning Based Approach for High Voltage Circuit Breaker Mechanical Fault Diagnosis

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
WW01 Invention patent application withdrawn after publication
WW01 Invention patent application withdrawn after publication

Application publication date: 20200804