CN114139591A - Mechanical equipment state early warning method - Google Patents
Mechanical equipment state early warning method Download PDFInfo
- Publication number
- CN114139591A CN114139591A CN202110694947.7A CN202110694947A CN114139591A CN 114139591 A CN114139591 A CN 114139591A CN 202110694947 A CN202110694947 A CN 202110694947A CN 114139591 A CN114139591 A CN 114139591A
- Authority
- CN
- China
- Prior art keywords
- data
- value
- mechanical equipment
- lstm
- early warning
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000012549 training Methods 0.000 claims abstract description 46
- 239000013598 vector Substances 0.000 claims description 22
- 238000000605 extraction Methods 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 5
- 238000000638 solvent extraction Methods 0.000 claims description 5
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000001514 detection method Methods 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 102100029469 WD repeat and HMG-box DNA-binding protein 1 Human genes 0.000 claims description 2
- 101710097421 WD repeat and HMG-box DNA-binding protein 1 Proteins 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims 1
- 238000012544 monitoring process Methods 0.000 abstract description 3
- 230000007547 defect Effects 0.000 abstract description 2
- 238000003745 diagnosis Methods 0.000 abstract description 2
- 238000004519 manufacturing process Methods 0.000 description 4
- 230000005856 abnormality Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
Images
Classifications
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The patent relates to the field of mechanical equipment state monitoring and fault diagnosis, in particular to a mechanical equipment state early warning method. The fixed threshold alarm method in the prior art has the defects that the data change trend of data cannot be sensed, and the change of the data change trend is often an early signal of fault occurrence. The method comprises five steps: the method comprises the following steps: extracting characteristics; step two: dividing data; step three: training a model; step four: self-learning of a threshold value; step five: and (5) state early warning. The invention provides an LSTM-based mechanical equipment state early warning method, which is characterized in that a prediction model is established for normal state data of mechanical equipment by using the LSTM, an alarm threshold value is learned by self, and an alarm is given when the difference between the prediction data and the real data exceeds the self-learning threshold value.
Description
Technical Field
The patent relates to the field of mechanical equipment state monitoring and fault diagnosis, in particular to a mechanical equipment state early warning method.
Background
Mechanical equipment plays an important role in modern industry, people put forward higher and higher requirements on state monitoring of the mechanical equipment along with the advance of industrial internet technology in recent years, abnormality is found at the early stage of the failure of the mechanical equipment, predictive maintenance is carried out, and the method plays an important role in reducing production cost and guaranteeing production safety of enterprises. However, in actual production, an alarm mode with a fixed threshold is still adopted, and when the alarm occurs, the fault is often worsened to a certain degree, so that great threat is brought to safe production, and an effective state early warning method needs to be developed. The fixed threshold alarm method has the defects that the data change trend of data cannot be sensed, and the change of the data change trend is often an early signal of failure, so that the accurate capture of the change of the data change trend is the key of the early warning method of mechanical equipment.
The LSTM is a specially designed Recurrent Neural Networks (RNN), solves the problem of long-term dependence of the traditional RNN, has good performance on time sequence data prediction, and is applied to the industries of language understanding, transportation and the like at present. Based on the self-learning alarm threshold value, the invention provides a mechanical equipment state early warning method based on LSTM, which comprises the steps of establishing a prediction model for normal state data of the mechanical equipment by using the LSTM, self-learning the alarm threshold value, and alarming when the difference between the predicted data and the real data exceeds the self-learning threshold value.
Disclosure of Invention
The invention aims to solve the problem that a fixed threshold alarm method cannot accurately give an early warning, and provides an LSTM-based mechanical equipment state early warning method.
A mechanical equipment state early warning method comprises five steps:
the method comprises the following steps: extracting characteristics, namely acquiring normal state data of the mechanical equipment, and extracting the characteristics of the normal state data to form a normal state characteristic data set of the mechanical equipment;
step two: data partitioning, dividing the normal state feature data set into XMAnd XTRespectively used for model construction and threshold self-learning;
step three: model training, establishing LSTM network structure, and utilizing XMTraining an LSTM, and taking the LSTM model after training as a prediction model;
step four: threshold value self-learning by XTAs test data of a prediction model, calculating the Mahalanobis distance between the predicted value and the true value of each group of test data, and calculating an alarm threshold value T by a box type graph method;
step five: and (3) state early warning, wherein the real-time vibration data of the mechanical equipment is subjected to feature extraction to be used as a real-time real value, a real-time predicted value is obtained based on data in a past period of time, the difference between the real-time real value and the real-time predicted value is calculated and compared with T, when the difference value is greater than T, an alarm is given, and if the difference value is greater than T, the alarm is not given.
The first step is as follows: the feature extraction specifically comprises the following steps: collecting normal state vibration data of mechanical equipment to be detected within a period of time, and dividing the normal state data into N groups of data Z ' ═ Z ' according to certain intervals ',(q)}q=1,...,NWherein Z',(q)And arranging the samples of each group in Z' from small to large according to the time sequence for the q-th group of data, and performing feature extraction on the normal state data of each group to obtain a feature vector set.
The second step is that: the data division specifically comprises the following steps: (1) predicting data of the current time by a prediction model based on the LSTM through a piece of historical data, recording the length of the historical data required by each prediction as L, wherein L belongs to [30,90], and (2) inputting the historical data with the length of L and outputting the historical data as a predicted value of the current time by the prediction model, and (3) dividing a normal state feature vector set Z to form a sample set of the prediction model.
The sample set is divided as follows: z contains N time-point feature vectors in total, from the L-th feature vector Z(L)Initially, the first and last eigenvectors of the L eigenvectors are respectively used as input data of the sample set and corresponding true values, and the sample composition of the sample set is expressed as D { (X)(r),Y(r))}r=1,…,N-L,X(r),Y(r)Data and corresponding true values of the r-th sample in the sample set respectively
Dividing a sample set D intoAndtwo moieties of which DMFor training early warning models, DTFor self-learning alarm thresholds.
The third step is that: model training, including two parts of LSTM network structure determination and prediction model training, wherein the parameters to be determined by the LSTM network structure include input data size, output data size, LSTM hidden layer neuron number and LSTM unit number; after the structure of the LSTM model is determined, D is adoptedMTraining the LSTM model, making the predicted value output by the LSTM model be H (X), adopting the mean square error as the objective function, and defining the objective function as formula 3
Wherein V is the number of samples per training batch, X(p)、Y(p)Respectively the data of the p-th sample in each batch of training samples and the corresponding real value.
And the prediction model training is to adopt a batch training method to minimize O to complete the training of the LSTM model, and the LSTM model after training can be used as a prediction model of the normal state data of the mechanical equipment to be monitored.
The input data size is the length L of historical data and the characteristic dimension S required by each prediction; the size of the output data is a predicted value of S characteristics in each prediction; the number of LSTM hidden layer neurons and the number of LSTM units are determined by combining actual data.
The fourth step is that: the threshold self-learning specifically comprises the following steps: will DTData of inner N-L-P samplesInputting the prediction model to obtain the predicted valueCalculating predicted value and true valueThe difference value between the predicted value and the true value is calculated through the Mahalanobis distance,the difference value is calculated as formula 4:
wherein Σ isThe superscript T and-1 of the covariance matrix respectively represent transposition operation and matrix inversion;
calculating DTThe difference value d of the inner N-L-P samples is { d ═ d(i)}i=N-L-PWhen d is arranged from small to large, the value at 25% is denoted as Q1The value at the top 75% is denoted as Q1According to the abnormal value detection method of the boxplot, the definition of the self-learning threshold value is shown as the formula 5
T=Q3+1.5×(Q3-Q1) (5)。
The fifth step is as follows: the state early warning specifically comprises:
acquiring state data of mechanical equipment to be monitored from time t-L to time t, and extracting S characteristics same in the step I to obtain a characteristic vector z ═ { z ═ z(t-L),z(t-L+1),…,z(t)Is at least one of
As an input of the prediction model, the output of the prediction model is represented as H [ X ](t)](ii) a Calculating the difference between the predicted value and the actual value of the time t according to equation 7, when d(t)If the alarm is greater than T, the alarm is given, otherwise, the alarm is not given
Drawings
FIG. 1 is a flow chart of a method for early warning of the state of a mechanical device based on LSTM according to the present invention;
FIG. 2 is a vibration waveform data segment of normal state data in an embodiment of the present invention;
FIG. 3 is a feature extraction result of normal state data in an embodiment of the present invention;
FIG. 4 is a graph comparing predicted values and actual values in an embodiment of the present invention;
FIG. 5 is a graph showing the variation of each characteristic of test data according to an embodiment of the present invention;
fig. 6 is an experimental result on test data in the example of the present invention.
Detailed Description
The invention provides an LSTM-based mechanical equipment state early warning method which is divided into two stages of model construction and model application, as shown in the attached figure 1, the method comprises the following five steps:
the method comprises the following steps: and (5) feature extraction. And acquiring normal state data of the mechanical equipment, and performing feature extraction on the normal state data to form a normal state feature data set of the mechanical equipment.
Step two: and (4) dividing data. Partitioning a Normal State feature dataset into XMAnd XTRespectively used for model construction and threshold self-learning.
Step three: and (5) training a model. Building LSTM network structure, using XMAnd training the LSTM, and taking the trained LSTM model as a prediction model.
Step four: the threshold value is self-learning. With XTAnd (3) as the test data of the prediction model, calculating the Mahalanobis distance between the predicted value and the true value of each group of test data, and calculating the alarm threshold value T by using a box type graph method.
Step five: and (5) state early warning. And (3) extracting the characteristics of the real-time vibration data of the mechanical equipment to obtain a real-time real value, obtaining a real-time predicted value based on the data of a period of time in the past, calculating the difference between the real-time real value and the real-time predicted value, comparing the difference with T, and giving an alarm if the difference is greater than T, or not giving an alarm if the difference is greater than T.
The five steps of the present invention will be described in detail below.
(1) The method comprises the following steps: feature extraction
Collecting normal state vibration data of mechanical equipment to be detected within a period of time, and dividing the normal state data into N groups of data Z ' ═ Z ' at certain intervals ',(q)}q=1,…,NWherein Z',(q)The q-th group of data is formed by arranging the samples in Z' from small to large in time sequence. Extracting features of each group of normal state data to obtain a feature vector set, for example, Z',(q)The feature vector obtained by extracting S features isWhereinIs Z',(q)The feature value of the jth feature, the normal state feature vector set obtained from Z' is Z ═ { Z ═ Z(q)}q=1,…,N。
(2) Step two: data partitioning
The prediction model based on LSTM predicts data at the current time from a single piece of history data, and records the length of the history data required for each prediction as L, generally L ∈ [30,90], so the input of the prediction model is the history data with length L, and the output is the predicted value at the current time. In order to complete the training of the prediction model, the normal state feature vector set Z needs to be divided to form a sample set of the prediction model, and the method for dividing the sample set is as follows:
z contains N time-point feature vectors in total, from the L-th feature vector Z(L)Initially, the first L eigenvectors (including the L-th eigenvector) and the next eigenvector are respectively used as input data of the sample set and corresponding true values, as shown in equations 1 and 2. The sample composition of the sample set may thus be denoted as D { (X)(r),Y(r))}r=1,…,N-L,X(r),Y(r)Respectively, the data of the r-th sample in the sample set and the corresponding true value.
In the construction method of the early warning model, normal state data are required to be respectively used for training of the prediction model and self-learning of the alarm threshold value, and a sample set D is divided intoAndtwo moieties of which DMFor training early warning models, DTFor self-learning alarm thresholds.
(3) Step three: model training
And step three, comprising two parts of structure determination of the LSTM network and training of a prediction model. The parameters to be determined by the LSTM network structure comprise input data size, output data size, LSTM hidden layer neuron number and LSTM unit number, and in the method, the specific meaning of the input data size is historical data length L and characteristic dimension S required by each prediction; the specific meaning of the output data size is the predicted value of S characteristics in each prediction; the number of neurons in the LSTM hidden layer and the number of LSTM units are related to the characteristics of actual data, the larger the number of the neurons and the LSTM units is, the stronger the fitting capability of the model is, but the training time and the overfitting possibility of the model are also increased, so that the determination needs to be performed by combining the actual data.
After the structure of the LSTM model is determined, D is adoptedMTraining the LSTM model, enabling a predicted value output by the LSTM model to be H (X), adopting a mean square error as an objective function, defining the objective function as formula 3, adopting a batch training method to minimize O to complete the training of the LSTM model, and enabling the trained LSTM model to be used as a prediction model of normal state data of the mechanical equipment to be monitored.
Wherein V is the number of samples per training batch, X(p)、Y(p)Respectively the data of the p-th sample in each batch of training samples and the corresponding real value.
(4) Step four: threshold self-learning
The prediction model has the capability of predicting the actual value at the current moment according to a section of historical data, when the deviation between the predicted value and the actual value of the prediction model is too large, the fact that the alarm needs to be given when the operation state of the mechanical equipment changes is indicated, and when the deviation between the predicted value and the actual value is large, the alarm is the key for judging whether the early warning model can accurately give an early warning or not. To overcome the defectThe invention self-learns the alarm early warning by normal state data and statistical method. Will DTData of inner N-L-P samplesInputting the prediction model to obtain the predicted valueCalculating predicted value and true valueThe difference value of (a). Because each real value comprises a plurality of characteristics and the scales of the characteristics are inconsistent, the difference value between the predicted value and the real value is calculated through the Mahalanobis distance so as toFor example, the difference value is calculated according to equation 4.
Wherein Σ isThe superscript T, -1 of the covariance matrix of (1) represents the transposition operation and matrix inversion, respectively.
Calculating DTThe difference value d of the inner N-L-P samples is { d ═ d(i)}i=N-L-PThe values of d from small to large are arranged and the value at 25% (lower quartile) is denoted as Q1And the value at the first 75% (lower quartile) is recorded as Q1According to the detection method of abnormal values of the boxed graph, the definition of the self-learning threshold value is shown as a formula 5.
T=Q3+1.5×(Q3-Q1) (5)
(5) Step five: state warning
Acquiring state data of mechanical equipment to be monitored from time t-L to time t, and extracting S characteristics identical to those of the first stepObtaining a feature vector z ═ z(t-L),z(t-L+1),…,z(t)Is at least one of
As an input of the prediction model, the output of the prediction model is represented as H [ X ](t)]. Calculating the difference between the predicted value and the actual value of the time t according to equation 7, when d(t)If the alarm is greater than T, the alarm is given, otherwise, the alarm is not given.
In order to describe the present invention more specifically, the following will take pump shaft vibration data of a main pump of a nuclear power plant as an example, and in conjunction with the accompanying drawings, describe an embodiment of the LSTM-based mechanical equipment state warning method in detail.
(1) The method comprises the following steps: feature extraction
Fig. 2 shows a vibration waveform diagram of a pump shaft of a nuclear power main pump in a normal state for 1 second, and the vibration waveform data of the pump shaft for 160 minutes in the normal state of the main pump is subjected to feature extraction, and a peak value, a half-frequency multiplication amplitude, a first-frequency multiplication amplitude, a second-frequency multiplication amplitude, a third-frequency multiplication amplitude, a fourth-frequency multiplication amplitude and a fifth-frequency multiplication amplitude are sequentially extracted to obtain 9600 groups of feature vectors, that is, N is 9600 and S is 7 in the content of the invention. The time-varying curves of the features are shown in fig. 3, Feature _1 to Feature _7 are respectively a peak value, a half-doubled frequency, a first-doubled frequency amplitude, a second-doubled frequency amplitude, a third-doubled frequency amplitude, a fourth-doubled frequency amplitude and a fifth-doubled frequency amplitude, and 7 Feature values corresponding to each time point in the graph form a set of Feature vectors.
(2) Step two: data partitioning
In this embodiment, the length L of the history data required for each prediction is 60, so that the input data and the corresponding true value of any sample in the sample set are shown in formulas 8 and 9, respectively, where 9540 ≧ r ≧ 1.
In the formulas 8 and 9, the first and second groups,the peak-to-peak value, half frequency multiplication, first frequency multiplication amplitude, second frequency multiplication amplitude, third frequency multiplication amplitude, fourth frequency multiplication amplitude and fifth frequency multiplication amplitude of the r group of characteristic vectors in the characteristic vector set are respectively.
This gave a 9540 sample set D { (X)(r),Y(r))}r=1,…,9540Dividing D into 1:1Andtwo moieties of which DMFor training early warning models, DTFor self-learning alarm thresholds.
(3) Step three: model training
The input data size is related to the length of the historical data and the characteristic dimension required by single prediction, so the input data size is 60 × 7 in the embodiment; the output data size is related to the feature dimension, which is 7 in this embodiment; in addition, in this embodiment, the number of LSTM units is 2, the number of hidden layer neurons is 128, and the data is an optimized value obtained through experiments.
By using DMTraining an LSTM model, wherein a predicted value output by the LSTM model is H (X), a mean square error is used as an objective function, the definition of the objective function is as shown in the formula 10, the training of the LSTM model is completed by minimizing O by adopting a batch training method, the number of samples trained in each batch is 1024, 2000 rounds of training are performed in total, and the trained LSTM model can be used as a prediction model of normal state data of mechanical equipment to be monitored, for example, FIG. 4 shows thatPredicted values (dashed lines) versus true values (solid lines) for the prediction model.
In the formula, X(p)、Y(p)∈DMThe data of the p-th sample in each batch of training samples and the corresponding true value are respectively.
(4) Step four: threshold self-learning
DMInThe covariance matrix of (A) is shown in formula 11. the quartile number Q can be obtained according to formula 4 in the invention12.901, upper quartile Q34.251, the alarm pre-threshold T is 4.251+1.5 × (4.251-2.901) ═ 6.276.
(5) Step five: state warning
Fig. 5 is a time-dependent change curve of 7 features of test data for verifying the effect of the present invention, in which the time at which the change rule of the features changes is marked with an ellipse, and it can be seen from the graph that the peak value, the half-frequency amplitude, the first-frequency amplitude, and the second-frequency amplitude slowly increase after 2000 time points.
And sequentially and slidingly intercepting the feature vectors of 60 time points of the test data as the input of a prediction model to calculate to obtain a predicted value of the next moment, and calculating the difference value between the predicted value and the true value according to the formula 7 in the invention content.
Claims (9)
1. A mechanical equipment state early warning method is characterized in that: the method comprises the following five steps:
the method comprises the following steps: extracting characteristics, namely acquiring normal state data of the mechanical equipment, and extracting the characteristics of the normal state data to form a normal state characteristic data set of the mechanical equipment;
step two: data partitioning, dividing the normal state feature data set into XMAnd XTRespectively used for model construction and threshold self-learning;
step three: model training, establishing LSTM network structure, and utilizing XMTraining an LSTM, and taking the LSTM model after training as a prediction model;
step four: threshold value self-learning by XTAs test data of a prediction model, calculating the Mahalanobis distance between the predicted value and the true value of each group of test data, and calculating an alarm threshold value T by a box type graph method;
step five: and (3) state early warning, wherein the real-time vibration data of the mechanical equipment is subjected to feature extraction to be used as a real-time real value, a real-time predicted value is obtained based on data in a past period of time, the difference between the real-time real value and the real-time predicted value is calculated and compared with T, when the difference value is greater than T, an alarm is given, and if the difference value is greater than T, the alarm is not given.
2. The mechanical equipment state early warning method according to claim 1, characterized in that: the method comprises the following steps: the feature extraction specifically comprises the following steps: collecting normal state vibration data of mechanical equipment to be detected within a period of time, and dividing the normal state data into N groups of data Z' ═ Z according to a certain interval',(q)}q=1,…,NWherein Z is',(q)And arranging the samples of each group in Z' from small to large according to the time sequence for the q-th group of data, and performing feature extraction on the normal state data of each group to obtain a feature vector set.
3. The mechanical equipment state early warning method according to claim 1, characterized in that: step two: the data division specifically comprises the following steps: the prediction model based on the LSTM predicts data at the current time through a piece of historical data, the length of the historical data required by each prediction is recorded as L, generally, L belongs to [30,90], the input of the prediction model is the historical data with the length of L, the output of the prediction model is the predicted value at the current time, and a normal state feature vector set Z is divided to form a sample set of the prediction model.
4. The mechanical equipment state early warning method according to claim 3, characterized in that: the sample set is divided as follows: z contains N time-point feature vectors in total, from the L-th feature vector Z(L)Initially, the first and last eigenvectors of the L eigenvectors are respectively used as input data and corresponding true values of the sample set, and the sample composition of the sample set is expressed as D { (X)(r),Y(r))}r=1,…,N-L,X(r),Y(r)Data and corresponding true values of the r-th sample in the sample set respectively
5. The mechanical equipment state early warning method according to claim 4, wherein: step three: model training, including two parts of LSTM network structure determination and prediction model training, wherein the parameters to be determined by the LSTM network structure include input data size, output data size, LSTM hidden layer neuron number and LSTM unit number; after the structure of the LSTM model is determined, D is adoptedMTraining the LSTM model, making the predicted value output by the LSTM model be H (X), adopting the mean square error as the objective function, and defining the objective function as formula 3
Wherein V is the number of samples per training batch, X(p)、Y(p)Respectively the data of the p-th sample in each batch of training samples and the corresponding real value.
6. The mechanical equipment state early warning method according to claim 5, wherein: and (3) training a prediction model, namely minimizing O by adopting a batch training method to finish the training of the LSTM model, wherein the LSTM model after being trained can be used as the prediction model of the normal state data of the mechanical equipment to be monitored.
7. The mechanical equipment state early warning method according to claim 4, wherein: the input data size is the length L of historical data and the characteristic dimension S required by each prediction; the size of the output data is a predicted value of S characteristics in each prediction; the number of neurons in the LSTM hidden layer and the number of LSTM units are determined according to actual data.
8. The mechanical equipment state early warning method according to claim 1, characterized in that: the fourth step is that: the threshold self-learning specifically comprises the following steps: will DTData of inner N-L-P samplesInputting the prediction model to obtain the predicted valueCalculating predicted value and true valueThe difference value between the predicted value and the true value is calculated through the Mahalanobis distance,the calculation method of the difference value is as shown in formula 4:
wherein Σ isThe superscript T and-1 of the covariance matrix respectively represent transposition operation and matrix inversion;
calculating DTThe difference value d of the inner N-L-P samples is { d ═ d(i)}i=N-L-PWhen d is arranged from small to large, the value at 25% is denoted as Q1The value at the top 75% is denoted as Q1According to the abnormal value detection method of the boxplot, the definition of the self-learning threshold value is shown as the formula 5
T=Q3+1.5×(Q3-Q1) (5)。
9. The mechanical equipment state early warning method according to claim 1, characterized in that: the fifth step is as follows: the state early warning specifically comprises:
acquiring state data of mechanical equipment to be monitored from time t-L to time t, and extracting S characteristics same in the step I to obtain a characteristic vector z ═ { z ═ z(t-L),z(t-L+1),…,z(t)Is at least one of
As an input of the prediction model, the output of the prediction model is represented as H [ X ](t)](ii) a Calculating the difference value between the predicted value and the true value at the moment t according to the formula 7, and calculating the difference value when d is(t)If the alarm is greater than T, the alarm is given, otherwise, the alarm is not given
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110694947.7A CN114139591A (en) | 2021-06-23 | 2021-06-23 | Mechanical equipment state early warning method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110694947.7A CN114139591A (en) | 2021-06-23 | 2021-06-23 | Mechanical equipment state early warning method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114139591A true CN114139591A (en) | 2022-03-04 |
Family
ID=80394092
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110694947.7A Pending CN114139591A (en) | 2021-06-23 | 2021-06-23 | Mechanical equipment state early warning method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114139591A (en) |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110008079A (en) * | 2018-12-25 | 2019-07-12 | 阿里巴巴集团控股有限公司 | Monitor control index method for detecting abnormality, model training method, device and equipment |
CN110298455A (en) * | 2019-06-28 | 2019-10-01 | 西安因联信息科技有限公司 | A kind of mechanical equipment fault intelligent early-warning method based on multivariable estimation prediction |
CN110763929A (en) * | 2019-08-08 | 2020-02-07 | 浙江大学 | Intelligent monitoring and early warning system and method for convertor station equipment |
CN111667020A (en) * | 2020-06-28 | 2020-09-15 | 中核武汉核电运行技术股份有限公司 | Fault early warning method for nuclear power rotating machinery |
CN112287602A (en) * | 2020-10-28 | 2021-01-29 | 北京国信会视科技有限公司 | Motor car axle temperature fault early warning method based on machine learning and isolated forest |
CN112380487A (en) * | 2020-11-02 | 2021-02-19 | 湖南江河机电自动化设备股份有限公司 | Method for judging running state of small hydraulic turbine set by using neural network self-coding |
US20210089900A1 (en) * | 2019-09-20 | 2021-03-25 | Wuhan University | Transformer dga data prediction method based on multi-dimensional time sequence frame convolution lstm |
CN112990435A (en) * | 2021-03-22 | 2021-06-18 | 华北电力大学 | Long-short-time memory network power station fan fault early warning method and system |
CN113011656A (en) * | 2021-03-22 | 2021-06-22 | 内蒙古电力(集团)有限责任公司内蒙古电力科学研究院分公司 | Power station auxiliary machine fault early warning method and system |
-
2021
- 2021-06-23 CN CN202110694947.7A patent/CN114139591A/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110008079A (en) * | 2018-12-25 | 2019-07-12 | 阿里巴巴集团控股有限公司 | Monitor control index method for detecting abnormality, model training method, device and equipment |
CN110298455A (en) * | 2019-06-28 | 2019-10-01 | 西安因联信息科技有限公司 | A kind of mechanical equipment fault intelligent early-warning method based on multivariable estimation prediction |
CN110763929A (en) * | 2019-08-08 | 2020-02-07 | 浙江大学 | Intelligent monitoring and early warning system and method for convertor station equipment |
US20210089900A1 (en) * | 2019-09-20 | 2021-03-25 | Wuhan University | Transformer dga data prediction method based on multi-dimensional time sequence frame convolution lstm |
CN111667020A (en) * | 2020-06-28 | 2020-09-15 | 中核武汉核电运行技术股份有限公司 | Fault early warning method for nuclear power rotating machinery |
CN112287602A (en) * | 2020-10-28 | 2021-01-29 | 北京国信会视科技有限公司 | Motor car axle temperature fault early warning method based on machine learning and isolated forest |
CN112380487A (en) * | 2020-11-02 | 2021-02-19 | 湖南江河机电自动化设备股份有限公司 | Method for judging running state of small hydraulic turbine set by using neural network self-coding |
CN112990435A (en) * | 2021-03-22 | 2021-06-18 | 华北电力大学 | Long-short-time memory network power station fan fault early warning method and system |
CN113011656A (en) * | 2021-03-22 | 2021-06-22 | 内蒙古电力(集团)有限责任公司内蒙古电力科学研究院分公司 | Power station auxiliary machine fault early warning method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110298455B (en) | Mechanical equipment fault intelligent early warning method based on multivariate estimation prediction | |
CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
Shi et al. | Rolling bearing initial fault detection using long short-term memory recurrent network | |
CN107145645B (en) | Method for predicting residual life of non-stationary degradation process with uncertain impact | |
Vachtsevanos et al. | Fault prognosis using dynamic wavelet neural networks | |
CN113723010A (en) | Bridge damage early warning method based on LSTM temperature-displacement correlation model | |
CN106768262B (en) | A kind of transformer online monitoring method based on surface vibration signals analysis | |
CN105868557A (en) | Online prediction method for remaining life of electromechanical equipment under situation of two-stage degradation | |
Zhang et al. | Remaining Useful Life Prediction of Rolling Bearings Using Electrostatic Monitoring Based on Two‐Stage Information Fusion Stochastic Filtering | |
CN114091525A (en) | Rolling bearing degradation trend prediction method | |
CN116292367B (en) | Power plant fan system abnormal condition detection method based on one-dimensional convolution | |
Qin et al. | Remaining useful life prediction for rotating machinery based on optimal degradation indicator | |
CN112598144A (en) | CNN-LSTM burst fault early warning method based on correlation analysis | |
Wei et al. | Remaining useful life estimation based on gamma process considered with measurement error | |
Tang et al. | Prediction of bearing performance degradation with bottleneck feature based on LSTM network | |
Senanayaka et al. | Autoencoders and recurrent neural networks based algorithm for prognosis of bearing life | |
Kizito et al. | Long short-term memory networks for facility infrastructure failure and remaining useful life prediction | |
CN116595319A (en) | Prediction method and system applied to rail transit motor health state evaluation | |
CN117575027B (en) | Sensor fault diagnosis method for enhancing space-time causality by using knowledge | |
CN117633690A (en) | Rotary machine health state monitoring method and equipment based on data driving | |
CN117473260A (en) | Method for predicting residual life of space bearing by digital-analog linkage | |
CN114139591A (en) | Mechanical equipment state early warning method | |
CN114925476B (en) | Rolling bearing degradation track augmentation quaternion prediction method and storage medium | |
Luo et al. | A novel method for remaining useful life prediction of roller bearings involving the discrepancy and similarity of degradation trajectories | |
Geramifard et al. | Continuous health assessment using a single hidden Markov model |
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 |