CN109814527A - Based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method and device - Google Patents
Based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method and device Download PDFInfo
- Publication number
- CN109814527A CN109814527A CN201910026193.0A CN201910026193A CN109814527A CN 109814527 A CN109814527 A CN 109814527A CN 201910026193 A CN201910026193 A CN 201910026193A CN 109814527 A CN109814527 A CN 109814527A
- Authority
- CN
- China
- Prior art keywords
- feature
- vector
- prediction
- neural network
- training
- 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.)
- Granted
Links
Abstract
The invention discloses one kind to be based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method and device, wherein, method is the following steps are included: obtain the Condition Monitoring Data collection of multiple sensors on target device periphery, wherein Condition Monitoring Data collection includes the monitoring data at 0 moment to current time;Concentrating selection from Condition Monitoring Data using feature selecting standard includes the predicted characteristics of preset failure information, wherein feature selecting standard includes correlation metric and monotonicity index;Predicted characteristics progress Feature Conversion is obtained into predicted characteristics vector;Single step failure predication, chromic trouble prediction and predicting residual useful life are carried out to target device according to predicted characteristics vector sum failure predication network model.This method can effectively avoid the precision of prediction as caused by unreasonable preset failure threshold value insufficient, and confidence interval can be provided under the occasion of single step performance prediction, the long-term forecast to equipment performance and remaining life may be implemented.
Description
Technical field
The present invention relates to the failure predication technical fields of data-driven, in particular to a kind of to be based on LSTM (Long Short-
Term Memory is shot and long term memory network) Recognition with Recurrent Neural Network industrial equipment failure prediction method and device.
Background technique
The equipment fault prediction technique of the relevant technologies, data-driven is based primarily upon statistical analysis, Bayesian network, SVM
(Support Vector Machine, support vector machines), HMM (Hidden Markov Model, hidden Markov model) and
NN (Neural Network, neural network) etc. is realized.Although above method achieves well under particular task, special scenes
As a result, but that there are precision of predictions is lower, Generalization Ability is insufficient, is difficult to or can not carry out asking for long-term (long-term) prediction
Topic, has difficulties when applied to follow-up works such as O&M policy optimizations.
Failure predication principal mode is that predicting residual useful life and equipment performance are predicted, it is known that information is usually with sequence data
Form provides (the monitoring data sequence of such as industrial equipment), generally achievees the purpose that failure predication by establishing prediction model,
The parameter of the prediction model can from the acquistion of sequence data middle school to.Therefore, failure predication problem can be regarded as a sequence
Problem concerning study.
In recent years, RNNs (Recurrent Neural Networks, Recognition with Recurrent Neural Network) is solving the problems, such as Sequence Learning
When embody absolute advantage, such as exact timing problem, language model, simple pen drawing, speech recognition etc., therefore pre- in failure
Survey field also receives very big concern.The existing failure prediction method based on RNN generally passes through two ways and realizes: it is a kind of with
T, the feature at t-T, t-2T... moment is as input, using the feature at t+T moment as output, to reach the mesh of performance prediction
, however as the increase of T, precision of prediction can be significantly reduced, and when T is smaller then loses the ability of guide maintenance decision, because
And it is difficult to carry out practical application;Another kind is using RNN as Feature Selection Model, general by default based on the feature of RNN input
Fault threshold and exponential model calculate remaining life, therefore advantage and characteristic of the RNN compared with other neural networks is
To making full use of.
Summary of the invention
The present invention is directed to solve at least some of the technical problems in related technologies.
For this purpose, an object of the present invention is to provide one kind to be based on LSTM Recognition with Recurrent Neural Network industrial equipment failure predication
Method, this method can effectively avoid the precision of prediction as caused by unreasonable preset failure threshold value insufficient, in single step performance prediction
Occasion under can provide confidence interval, the long-term forecast to equipment performance and remaining life may be implemented.
It is a kind of based on LSTM Recognition with Recurrent Neural Network industrial equipment failure predication dress it is another object of the present invention to propose
It sets.
In order to achieve the above objectives, one aspect of the present invention embodiment is proposed one kind and is set based on the industry of LSTM Recognition with Recurrent Neural Network
Standby failure prediction method, comprising the following steps: step S101: the status monitoring number of multiple sensors on target device periphery is obtained
According to collection, wherein the Condition Monitoring Data collection includes the monitoring data at 0 moment to current time;Step S102: it is selected using feature
Selecting standard and concentrating selection from the Condition Monitoring Data includes the predicted characteristics of preset failure information, wherein the feature selecting
Standard includes correlation metric and monotonicity index;Step S103: it is special that predicted characteristics progress Feature Conversion is obtained into prediction
Levy vector;Step S104: single step is carried out to the target device according to the predicted characteristics vector sum failure predication network model
Failure predication, chromic trouble prediction and predicting residual useful life.
The embodiment of the present invention based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method, can effectively avoid by
Precision of prediction caused by unreasonable preset failure threshold value is insufficient;Obtaining performance prediction by way of Gaussian Mixture profile samples
Value, output be Gaussian Mixture distribution parameter, not only available current predicted value, can also obtain the distribution of predicted value
Situation can provide confidence interval under the occasion of single step performance prediction;The machine learning such as support vector machines and k neighbour are not depended on
The long-term forecast to equipment performance and remaining life may be implemented in method.
In addition, according to the above embodiment of the present invention be based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method also
It can have following additional technical characteristic:
Further, in one embodiment of the invention, the monitoring data include temperature data, pressure data and shape
One of variation data are a variety of.
Further, in one embodiment of the invention, the feature selecting standard are as follows:
Criteria=α Corr+ (1- α) Mon,
Wherein, α ∈ [0,1] is balance factor, and Corr is the correlation metric, and Mon is the monotonicity index;
Wherein, ith feature sequence is f(i), the time span of observation sequence is T, ft (i)It is i-th dimension feature in t moment
Sampling,For mean value, df(i)For f(i)Derivative.
Further, in one embodiment of the invention, it is described by the predicted characteristics progress Feature Conversion obtain it is pre-
Feature vector is surveyed, further comprises: being encoded using multiple equipment state of the one-hot vector to the target device, with
Obtain status indicator vector;Sensor monitoring vector is obtained according to the predicted characteristics, and vector is monitored according to the sensor
The predicted characteristics vector is obtained in the opposite variation of different moments and the status indicator vector.
Further, in one embodiment of the invention, the training method of the failure predication network model includes: to obtain
Take the monitoring history data set of multiple sensors on target device periphery;The monitoring is gone through according to step S102 and step S103
History data set is handled to obtain training feature vector;Training characteristics data set is obtained according to the monitoring history data set, and
Training pattern is built based on shot and long term memory unit and gauss hybrid models;It is special according to the training feature vector and the training
Sign data set is trained the training pattern, to obtain the failure predication network model.
In order to achieve the above objectives, another aspect of the present invention embodiment proposes a kind of based on the industry of LSTM Recognition with Recurrent Neural Network
Equipment fault prediction meanss, comprising: data acquisition module, the status monitoring of multiple sensors for obtaining target device periphery
Data set, wherein the Condition Monitoring Data collection includes the monitoring data at 0 moment to current time;Feature selection module is used for
Concentrating selection from the Condition Monitoring Data using feature selecting standard includes the predicted characteristics of preset failure information, wherein institute
Stating feature selecting standard includes correlation metric and monotonicity index;Feature Conversion module, for carrying out the predicted characteristics
Feature Conversion obtains predicted characteristics vector;Prediction module, for according to the predicted characteristics vector sum failure predication network model
Single step failure predication, chromic trouble prediction and predicting residual useful life are carried out to the target device.
The embodiment of the present invention based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device, can effectively avoid by
Precision of prediction caused by unreasonable preset failure threshold value is insufficient;Obtaining performance prediction by way of Gaussian Mixture profile samples
Value, output be Gaussian Mixture distribution parameter, not only available current predicted value, can also obtain the distribution of predicted value
Situation can provide confidence interval under the occasion of single step performance prediction;The machine learning such as support vector machines and k neighbour are not depended on
The long-term forecast to equipment performance and remaining life may be implemented in method.
In addition, according to the above embodiment of the present invention be based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device also
It can have following additional technical characteristic:
Further, in one embodiment of the invention, the monitoring data include temperature data, pressure data and shape
One of variation data are a variety of.
Further, in one embodiment of the invention, the feature selecting standard are as follows:
Criteria=α Corr+ (1- α) Mon,
Wherein, α ∈ [0,1] is balance factor, and Corr is the correlation metric, and Mon is the monotonicity index;
Wherein, ith feature sequence is f(i), the time span of observation sequence is T, ft (i)It is i-th dimension feature in t moment
Sampling,For mean value, df(i)For f(i)Derivative.
Further, in one embodiment of the invention, the Feature Conversion module is further used for utilizing one-hot
Vector encodes the multiple equipment state of the target device, to obtain status indicator vector, and it is special according to the prediction
Obtain sensor monitoring vector, and according to sensor monitor vector different moments opposite variation and the status indicator to
Measure the predicted characteristics vector.
Further, in one embodiment of the invention, further includes: model training module, for obtaining target device
The monitoring history data set of multiple sensors on periphery, and according to the feature selection module and the Feature Conversion module to institute
It states monitoring history data set to be handled to obtain training feature vector, training characteristics number is obtained according to the monitoring history data set
Training pattern is built according to collection, and based on shot and long term memory unit and gauss hybrid models, according to the training feature vector and institute
It states training characteristics data set to be trained the training pattern, to obtain the failure predication network model.
The additional aspect of the present invention and advantage will be set forth in part in the description, and will partially become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Above-mentioned and/or additional aspect and advantage of the invention will become from the following description of the accompanying drawings of embodiments
Obviously and it is readily appreciated that, in which:
Fig. 1 is according to one embodiment of the invention based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method
Flow chart;
Fig. 2 is according to one specific embodiment of the present invention based on LSTM Recognition with Recurrent Neural Network industrial equipment failure predication side
The flow chart of method;
Fig. 3 is the LSTM neuronal structure schematic diagram according to one embodiment of the invention;
Fig. 4 is the network model schematic diagram according to one embodiment of the invention;
Fig. 5 is according to one embodiment of the invention based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device
Structural schematic diagram.
Specific embodiment
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, it is intended to is used to explain the present invention, and is not considered as limiting the invention.
It describes to propose according to embodiments of the present invention with reference to the accompanying drawings former based on LSTM Recognition with Recurrent Neural Network industrial equipment
Hinder prediction technique and device, describe to propose according to embodiments of the present invention first with reference to the accompanying drawings based on LSTM Recognition with Recurrent Neural Network
Industrial equipment failure prediction method.
Fig. 1 is the process based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method of one embodiment of the invention
Figure.
As shown in Figure 1, should based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method the following steps are included:
Step S101: the Condition Monitoring Data collection of multiple sensors on target device periphery is obtained, wherein status monitoring number
It include the monitoring data at 0 moment to current time according to collection.Wherein, monitoring data include temperature data, pressure data and deformability
One of data are a variety of.
It is understood that as shown in Fig. 2, the embodiment of the present invention is passed by the pressure being deployed in around target component first
Sensor, temperature sensor and other type sensors etc. obtain historical datas and the real-time monitoring numbers such as temperature, pressure, deformability
According to.
Step S102: concentrating selection from Condition Monitoring Data using feature selecting standard includes the prediction of preset failure information
Feature, wherein feature selecting standard includes correlation metric and monotonicity index.
It is understood that the predicted characteristics comprising preset failure information can be the feature of most fault messages, such as Fig. 2
It is shown, feature selecting.The purpose of feature selecting is to pick out the feature comprising most fault messages.Since good feature is usual
Degenerative process with equipment be it is consistent and dull, therefore, the embodiment of the present invention using monotonicity index and coincident indicator into
The operation of row feature selecting.Wherein, coincident indicator features the linear relationship between feature and operation hours, and monotonicity refers to
Mark features the raising and lowering trend of feature value.
Wherein, in one embodiment of the invention, feature selecting standard are as follows:
Criteria=α Corr+ (1- α) Mon,
Wherein, α ∈ [0,1] is balance factor, and Corr is correlation metric, and Mon is monotonicity index;
Wherein, ith feature sequence is f(i), the time span of observation sequence is T, ft (i)It is i-th dimension feature in t moment
Sampling,For mean value, df(i)For f(i)Derivative.
Specifically, note, the sequence of observations acquired by i-th of sensor, i.e. ith feature sequence are f(i), observe sequence
The time span of column is T, ft (i)For i-th dimension feature t moment sampling,For mean value, df(i)For f(i)Derivative, then have:
Since susceptibility of the feature to failure is positively correlated with above-mentioned two index, to linear group of Corr and Mon
Conjunction can be used as feature selecting standard, and note α ∈ [0,1] is balance factor, have:
Criteria=α Corr+ (1- α) Mon.
Step S103: predicted characteristics progress Feature Conversion is obtained into predicted characteristics vector.
Specifically, as shown in Fig. 2, Feature Conversion.Assuming that the characteristic dimension after feature selecting operation is k, after note conversion
Feature be S, then the dimension of S be k+2, the value S of t momentt=(Δ ft, m1, m2), wherein Δ ft=ft-ft-1, t ∈ [1,
T], f0=f1, Δ ft∈Rk。(m1, m2) it is one-hot vector, represent two kinds of equipment running status: healthy, unhealthy.It is specific next
It says, m1=1 expression equipment is in health status, on the contrary, m2=1 indicates that failure occurs in equipment.
Step S104: it is pre- that single step failure is carried out to target device according to predicted characteristics vector sum failure predication network model
It surveys, chromic trouble is predicted and predicting residual useful life.
It is understood that as shown in figure 3, failure predication.T at any time, based on the model obtained by model training
Parameter, carry out the Single-step Prediction (One-stepPrediction) of failure, long-term forecast (Long-term Prediction) and
Predicting residual useful life (Remaining Useful Life Prediction).
Specifically, single step failure predication and the pseudocode of chromic trouble prediction are as shown in Table 1 and Table 2.tpThe residue longevity at moment
The step of life prediction are as follows: the continuous sampling since current timeUntilMoment state instruction to
AmountThen tpThe predicting residual useful life value at moment are as follows:
Table 1
Table 2
Further, in one embodiment of the invention, the training method of failure predication network model includes: acquisition mesh
The monitoring history data set of multiple sensors on marking device periphery;According to step S102 and step S103 to monitoring history data set
It is handled to obtain training feature vector;Training characteristics data set is obtained according to monitoring history data set, and is remembered based on shot and long term
Recall unit and gauss hybrid models build training pattern;According to training feature vector and training characteristics data set to training pattern into
Row training, to obtain failure predication network model.
It is understood that before carrying out model training, by the step of executing step S101 to step S103, i.e., such as Fig. 2
It is shown, identical processing is carried out to the data of acquisition, sensor historic data is obtained, then carries out feature selecting, then carry out spy
Sign conversion, and by the prediction model after model training.The detailed process of model training includes:
Specifically, the network structure of model is as shown in figure 4, be based on shot and long term memory unit (Long Short Term
Memory, LSTM) and gauss hybrid models (Gaussians Mixture Model, GMM) build.
Specifically, t at every point of time, LSTM network (neuronal structure is as shown in Figure 3) is with feature StAnd last moment
Hidden state ht-1For input, connected layer (Fully Connected Layer) output subsequent time feature S entirelyt+1Probability point
The variable of cloth, here using the gauss hybrid models being made of M normal distribution, it is assumed that μiAnd σiRespectively i-th of Gaussian Profile
Mean value and standard deviation, when fault signature dimension be 1 when, have:
Wherein,In addition, this example is using classification distribution (Categorical Distribution) (p1,
p2) vector (m is indicated to the operating status of equipment1, m2) modeled, meet p1+p2=1.By the combination of LSTM and GMM, originally
The model that technology is proposed can not only predict the characteristic value of following instant, while the predicted value of later moment in time can also be provided
Distribution.
It is respectively h that initial hidden state and input feature vector, which is arranged,0=0, S0=(0,1,0), ytIt can be split as the portion M+1
Point, M normal distribution and 1 category distribution are respectively corresponded, at this point, prediction model can indicate are as follows:
ht=LSTM (St-1, ht-1)
yt=Wyht+by
Since normal distribution standard difference has nonnegativity, this technology is using exp operator and softmax operator to probability
Distribution parameter proceeds as follows:
The embodiment of the present invention carries out model training, reconstruct by minimizing reconstructed error (ReconstructionError)
The output that error can connect layer by training set and entirely is calculated, loss function LRIt is as follows:
Loss=LR=Ls+Lp
Below will by specific embodiment to based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method carry out into
One step illustrates.
Further, in one embodiment of the invention, by predicted characteristics progress Feature Conversion obtain predicted characteristics to
Amount, is further comprised: being encoded using multiple equipment state of the one-hot vector to target device, to obtain status indicator
Vector;Sensor monitoring vector is obtained according to predicted characteristics, and vector is monitored in the opposite variation of different moments according to sensor
Predicted characteristics vector is obtained with status indicator vector.
Specifically, as shown in Fig. 2, the failure prediction method of the embodiment of the present invention specifically includes:
The line next stage: the sensor states monitoring history data set of target device is obtained;Utilize correlation metric (Corr
Metric) select to include the most k sensor monitor value of fault message with monotonicity index (Mon metric) as feature,
Sensor monitors vector f=[f1, f2..., fk];L equipment state is encoded using one-hot vector, obtains state
Mark vector m=[m1, m2..., ml];By variation and the status indicator vector relatively between the different moments of sensor monitoring vector
Merge, constitutes final feature vector, by taking t moment as an example, feature vector are as follows: st=[Δ ft, m], wherein Δ ft=ft-ft-1;
Training characteristics data set is obtained by historical dataWherein,It constructs as shown in Figure 3
Failure predication network model M, wherein the structure of LSTM neuron is as shown in Figure 2;It is input with S, is instructed using method end to end
Practice M.
On-line stage: processing identical with the line next stage is executed to real-time device sensor monitoring data, obtained for 0 moment
To the feature vector S at current time (by taking t moment as an example)1:t={ s1, s2..., st};Respectively according to as shown in Table 1 and Table 2
Process carries out a step performance prediction and long-term behaviour is predicted;It is surplus that equipment is calculated according to the result of long-term behaviour assessment and current time
The remaining service life.
To sum up, the embodiment of the present invention has following advantage:
1) it is not necessarily to preset failure threshold value.The method of the embodiment of the present invention is by the discretization working condition of industrial equipment with one-
The form of hot vector is inputted as part, thus the downtime of facility for study, and is exported in the form of one-hot vector, because
This can effectively be avoided the precision of prediction as caused by unreasonable preset failure threshold value insufficient without presetting fault threshold.
2) confidence interval can be provided under the occasion of single step performance prediction.With direct output equipment performance or remaining life
The way of predicted value is different, and the method for the embodiment of the present invention is obtaining performance prediction by way of Gaussian Mixture profile samples
Value, output be Gaussian Mixture distribution parameter, therefore this method not only available current predicted value can also obtain pre-
The distribution situation of measured value.
3) it is able to carry out long-term (long-term) performance prediction.The method of the embodiment of the present invention does not depend on support vector machines
(SVM) and the machine learning methods such as k neighbour (KNN) long-term forecast (long- to equipment performance and remaining life, may be implemented
term prediction)。
It is proposed according to embodiments of the present invention based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method, Neng Gouyou
Effect avoids the precision of prediction as caused by unreasonable preset failure threshold value insufficient;It is obtained by way of Gaussian Mixture profile samples
Performance prediction value, output be Gaussian Mixture distribution parameter, not only available current predicted value, can also be predicted
The distribution situation of value can provide confidence interval under the occasion of single step performance prediction;Support vector machines and k neighbour etc. are not depended on
The long-term forecast to equipment performance and remaining life may be implemented in machine learning method.
It is proposed according to embodiments of the present invention referring next to attached drawing description former based on LSTM Recognition with Recurrent Neural Network industrial equipment
Hinder prediction meanss.
Fig. 5 is the structure based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device of one embodiment of the invention
Schematic diagram.
As shown in figure 5, should include: data acquisition mould based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device 10
Block 100, feature selection module 200, Feature Conversion module 300 and prediction module 400.
Wherein, data acquisition module 100 is used to obtain the Condition Monitoring Data collection of multiple sensors on target device periphery,
Wherein, Condition Monitoring Data collection includes the monitoring data at 0 moment to current time.Feature selection module 200 is used to utilize feature
It includes the predicted characteristics of preset failure information that selection criteria, which concentrates selection from Condition Monitoring Data, wherein feature selecting standard packet
Include correlation metric and monotonicity index.Feature Conversion module 300 is used to predicted characteristics progress Feature Conversion obtaining prediction special
Levy vector.Prediction module 400 is used to carry out single step event to target device according to predicted characteristics vector sum failure predication network model
Barrier prediction, chromic trouble prediction and predicting residual useful life.The device 10 of the embodiment of the present invention can be avoided effectively by unreasonable pre-
If precision of prediction caused by fault threshold is insufficient, confidence interval can be provided under the occasion of single step performance prediction, may be implemented
Long-term forecast to equipment performance and remaining life.
Further, in one embodiment of the invention, monitoring data include temperature data, pressure data and deformability
One of data are a variety of.
Further, in one embodiment of the invention, feature selecting standard are as follows:
Criteria=α Corr+ (1- α) Mon,
Wherein, α ∈ [0,1] is balance factor, and Corr is correlation metric, and Mon is monotonicity index;
Wherein, ith feature sequence is f(i), the time span of observation sequence is T, ft (i)It is i-th dimension feature in t moment
Sampling,For mean value, df(i)For f(i)Derivative.
Further, in one embodiment of the invention, Feature Conversion module is further used for utilizing one-hot vector
The multiple equipment state of target device is encoded, to obtain status indicator vector, and obtains sensor according to predicted characteristics
Monitor vector, and according to sensor monitor vector the opposite variation of different moments and status indicator vector obtain predicted characteristics to
Amount.
Further, in one embodiment of the invention, the device of the embodiment of the present invention further include: model training mould
Block.Wherein, model training module is used to obtain the monitoring history data set of multiple sensors on target device periphery, and according to spy
Sign 300 pairs of monitoring history data sets of selecting module 200 and Feature Conversion module are handled to obtain training feature vector, according to prison
It surveys history data set and obtains training characteristics data set, and trained mould is built based on shot and long term memory unit and gauss hybrid models
Type is trained training pattern according to training feature vector and training characteristics data set, to obtain failure predication network model.
It should be noted that aforementioned to the solution based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method embodiment
Release explanation be also applied for the embodiment based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device, it is no longer superfluous herein
It states.
It is proposed according to embodiments of the present invention based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device, Neng Gouyou
Effect avoids the precision of prediction as caused by unreasonable preset failure threshold value insufficient;It is obtained by way of Gaussian Mixture profile samples
Performance prediction value, output be Gaussian Mixture distribution parameter, not only available current predicted value, can also be predicted
The distribution situation of value can provide confidence interval under the occasion of single step performance prediction;Support vector machines and k neighbour etc. are not depended on
The long-term forecast to equipment performance and remaining life may be implemented in machine learning method.
In addition, term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance
Or implicitly indicate the quantity of indicated technical characteristic.Define " first " as a result, the feature of " second " can be expressed or
Implicitly include at least one this feature.In the description of the present invention, the meaning of " plurality " is at least two, such as two, three
It is a etc., unless otherwise specifically defined.
In the present invention unless specifically defined or limited otherwise, fisrt feature in the second feature " on " or " down " can be with
It is that the first and second features directly contact or the first and second features pass through intermediary mediate contact.Moreover, fisrt feature exists
Second feature " on ", " top " and " above " but fisrt feature be directly above or diagonally above the second feature, or be merely representative of
First feature horizontal height is higher than second feature.Fisrt feature can be under the second feature " below ", " below " and " below "
One feature is directly under or diagonally below the second feature, or is merely representative of first feature horizontal height less than second feature.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
It must be directed to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be in office
It can be combined in any suitable manner in one or more embodiment or examples.In addition, without conflicting with each other, the skill of this field
Art personnel can tie the feature of different embodiments or examples described in this specification and different embodiments or examples
It closes and combines.
Although the embodiments of the present invention has been shown and described above, it is to be understood that above-described embodiment is example
Property, it is not considered as limiting the invention, those skilled in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, modifies, replacement and variant.
Claims (10)
1. one kind is based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method, which comprises the following steps:
Step S101: the Condition Monitoring Data collection of multiple sensors on target device periphery is obtained, wherein the status monitoring number
It include the monitoring data at 0 moment to current time according to collection;
Step S102: concentrating selection from the Condition Monitoring Data using feature selecting standard includes the prediction of preset failure information
Feature, wherein the feature selecting standard includes correlation metric and monotonicity index;
Step S103: predicted characteristics progress Feature Conversion is obtained into predicted characteristics vector;
Step S104: single step failure is carried out to the target device according to the predicted characteristics vector sum failure predication network model
Prediction, chromic trouble prediction and predicting residual useful life.
2. according to claim 1 be based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method, which is characterized in that
The monitoring data include one of temperature data, pressure data and deformability data or a variety of.
3. according to claim 1 be based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method, which is characterized in that
The feature selecting standard are as follows:
Criteria=α Corr+ (1- α) Mon,
Wherein, α ∈ [0,1] is balance factor, and Corr is the correlation metric, and Mon is the monotonicity index;
Wherein, ith feature sequence is f(i), the time span of observation sequence is T,For i-th dimension feature adopting in t moment
Sample,For mean value, df(i)For f(i)Derivative.
4. according to claim 1 be based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method, which is characterized in that
It is described that predicted characteristics progress Feature Conversion is obtained into predicted characteristics vector, further comprise:
It is encoded using multiple equipment state of the one-hot vector to the target device, to obtain status indicator vector;
Sensor monitoring vector is obtained according to the predicted characteristics, and vector is monitored in the phase of different moments according to the sensor
The predicted characteristics vector is obtained to variation and the status indicator vector.
5. according to claim 1-4 be based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method,
It is characterized in that, the training method of the failure predication network model includes:
Obtain the monitoring history data set of multiple sensors on target device periphery;
The monitoring history data set is handled to obtain training feature vector according to step S102 and step S103;
Training characteristics data set is obtained according to the monitoring history data set, and is based on shot and long term memory unit and Gaussian Mixture mould
Type builds training pattern;
The training pattern is trained according to the training feature vector and the training characteristics data set, it is described to obtain
Failure predication network model.
6. one kind is based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device characterized by comprising
Data acquisition module, the Condition Monitoring Data collection of multiple sensors for obtaining target device periphery, wherein the shape
State monitoring data collection includes the monitoring data at 0 moment to current time;
Feature selection module, for concentrating selection to believe comprising preset failure from the Condition Monitoring Data using feature selecting standard
The predicted characteristics of breath, wherein the feature selecting standard includes correlation metric and monotonicity index;
Feature Conversion module, for predicted characteristics progress Feature Conversion to be obtained predicted characteristics vector;
Prediction module, for carrying out single step to the target device according to the predicted characteristics vector sum failure predication network model
Failure predication, chromic trouble prediction and predicting residual useful life.
7. according to claim 6 be based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device, which is characterized in that
The monitoring data include one of temperature data, pressure data and deformability data or a variety of.
8. according to claim 6 be based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device, which is characterized in that
The feature selecting standard are as follows:
Criteria=α Corr+ (1- α) Mon,
Wherein, α ∈ [0,1] is balance factor, and Corr is the correlation metric, and Mon is the monotonicity index;
Wherein, ith feature sequence is f(i), the time span of observation sequence is T,For i-th dimension feature adopting in t moment
Sample,For mean value, df(i)For f(i)Derivative.
9. according to claim 6 be based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device, which is characterized in that
The Feature Conversion module is further used for compiling using multiple equipment state of the one-hot vector to the target device
Code to obtain status indicator vector, and obtains sensor monitoring vector according to the predicted characteristics, and according to sensor monitor to
It measures and obtains the predicted characteristics vector in the opposite variation of different moments and the status indicator vector.
10. it is based on LSTM Recognition with Recurrent Neural Network industrial equipment fault prediction device according to claim 6-9 is described in any item,
It is characterized in that, further includes:
Model training module, the monitoring history data set of multiple sensors for obtaining target device periphery, and according to described
Feature selection module and the Feature Conversion module are handled to obtain training feature vector, root to the monitoring history data set
Training characteristics data set is obtained according to the monitoring history data set, and is built based on shot and long term memory unit and gauss hybrid models
Training pattern is trained the training pattern according to the training feature vector and the training characteristics data set, with
To the failure predication network model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910026193.0A CN109814527B (en) | 2019-01-11 | 2019-01-11 | Industrial equipment fault prediction method and device based on LSTM recurrent neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910026193.0A CN109814527B (en) | 2019-01-11 | 2019-01-11 | Industrial equipment fault prediction method and device based on LSTM recurrent neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109814527A true CN109814527A (en) | 2019-05-28 |
CN109814527B CN109814527B (en) | 2020-11-13 |
Family
ID=66603342
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910026193.0A Active CN109814527B (en) | 2019-01-11 | 2019-01-11 | Industrial equipment fault prediction method and device based on LSTM recurrent neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109814527B (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110287583A (en) * | 2019-06-21 | 2019-09-27 | 上海交通大学 | Industrial equipment method for predicting residual useful life based on Recognition with Recurrent Neural Network |
CN110702418A (en) * | 2019-10-10 | 2020-01-17 | 山东超越数控电子股份有限公司 | Aircraft engine fault prediction method |
CN110764065A (en) * | 2019-10-16 | 2020-02-07 | 清华大学 | Radar fault diagnosis method based on time sequence reconstruction |
CN111191838A (en) * | 2019-12-27 | 2020-05-22 | 赛腾机电科技(常州)有限公司 | Industrial equipment state control method and device integrating artificial intelligence algorithm |
CN111240282A (en) * | 2019-12-31 | 2020-06-05 | 联想(北京)有限公司 | Process optimization method, device, equipment and computer readable storage medium |
CN111258302A (en) * | 2020-01-23 | 2020-06-09 | 北京航天自动控制研究所 | Aircraft thrust fault online identification method based on LSTM neural network |
CN111277444A (en) * | 2020-02-05 | 2020-06-12 | 苏州浪潮智能科技有限公司 | Switch fault early warning method and device |
CN111523647A (en) * | 2020-04-26 | 2020-08-11 | 南开大学 | Network model training method and device, and feature selection model, method and device |
CN111708682A (en) * | 2020-06-17 | 2020-09-25 | 腾讯科技(深圳)有限公司 | Data prediction method, device, equipment and storage medium |
CN111860569A (en) * | 2020-06-01 | 2020-10-30 | 国网浙江省电力有限公司宁波供电公司 | Power equipment abnormity detection system and method based on artificial intelligence |
CN112395806A (en) * | 2020-11-11 | 2021-02-23 | 北京京航计算通讯研究所 | Method and device for predicting residual life of comprehensive transmission hydraulic system |
CN112418306A (en) * | 2020-11-20 | 2021-02-26 | 上海工业自动化仪表研究院有限公司 | Gas turbine compressor fault early warning method based on LSTM-SVM |
CN113076913A (en) * | 2021-04-16 | 2021-07-06 | 嘉兴毕格智能科技有限公司 | Aircraft engine fault prediction method |
CN113240099A (en) * | 2021-07-09 | 2021-08-10 | 北京博华信智科技股份有限公司 | LSTM-based rotating machine health state prediction method and device |
CN113657628A (en) * | 2021-08-20 | 2021-11-16 | 武汉霖汐科技有限公司 | Industrial equipment monitoring method and system, electronic equipment and storage medium |
CN113837477A (en) * | 2021-09-27 | 2021-12-24 | 西安交通大学 | Data dual-drive power grid fault prediction method, device and equipment under typhoon disaster |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106603293A (en) * | 2016-12-20 | 2017-04-26 | 南京邮电大学 | Network fault diagnosis method based on deep learning in virtual network environment |
CN107769972A (en) * | 2017-10-25 | 2018-03-06 | 武汉大学 | A kind of power telecom network equipment fault Forecasting Methodology based on improved LSTM |
CN108037378A (en) * | 2017-10-26 | 2018-05-15 | 上海交通大学 | Running state of transformer Forecasting Methodology and system based on long memory network in short-term |
CN108344564A (en) * | 2017-12-25 | 2018-07-31 | 北京信息科技大学 | A kind of state recognition of main shaft features Testbed and prediction technique based on deep learning |
CN108536123A (en) * | 2018-03-26 | 2018-09-14 | 北京交通大学 | The method for diagnosing faults of the train control on board equipment of the long neural network of memory network combination in short-term |
-
2019
- 2019-01-11 CN CN201910026193.0A patent/CN109814527B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106603293A (en) * | 2016-12-20 | 2017-04-26 | 南京邮电大学 | Network fault diagnosis method based on deep learning in virtual network environment |
CN107769972A (en) * | 2017-10-25 | 2018-03-06 | 武汉大学 | A kind of power telecom network equipment fault Forecasting Methodology based on improved LSTM |
CN108037378A (en) * | 2017-10-26 | 2018-05-15 | 上海交通大学 | Running state of transformer Forecasting Methodology and system based on long memory network in short-term |
CN108344564A (en) * | 2017-12-25 | 2018-07-31 | 北京信息科技大学 | A kind of state recognition of main shaft features Testbed and prediction technique based on deep learning |
CN108536123A (en) * | 2018-03-26 | 2018-09-14 | 北京交通大学 | The method for diagnosing faults of the train control on board equipment of the long neural network of memory network combination in short-term |
Cited By (26)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110287583A (en) * | 2019-06-21 | 2019-09-27 | 上海交通大学 | Industrial equipment method for predicting residual useful life based on Recognition with Recurrent Neural Network |
CN110702418A (en) * | 2019-10-10 | 2020-01-17 | 山东超越数控电子股份有限公司 | Aircraft engine fault prediction method |
CN110764065A (en) * | 2019-10-16 | 2020-02-07 | 清华大学 | Radar fault diagnosis method based on time sequence reconstruction |
CN110764065B (en) * | 2019-10-16 | 2021-10-08 | 清华大学 | Radar fault diagnosis method based on time sequence reconstruction |
CN111191838A (en) * | 2019-12-27 | 2020-05-22 | 赛腾机电科技(常州)有限公司 | Industrial equipment state control method and device integrating artificial intelligence algorithm |
CN111191838B (en) * | 2019-12-27 | 2023-09-22 | 赛腾机电科技(常州)有限公司 | Industrial equipment state management and control method and device integrating artificial intelligence algorithm |
CN111240282A (en) * | 2019-12-31 | 2020-06-05 | 联想(北京)有限公司 | Process optimization method, device, equipment and computer readable storage medium |
CN111240282B (en) * | 2019-12-31 | 2021-12-24 | 联想(北京)有限公司 | Process optimization method, device, equipment and computer readable storage medium |
CN111258302A (en) * | 2020-01-23 | 2020-06-09 | 北京航天自动控制研究所 | Aircraft thrust fault online identification method based on LSTM neural network |
CN111258302B (en) * | 2020-01-23 | 2021-10-01 | 北京航天自动控制研究所 | Aircraft thrust fault online identification method based on LSTM neural network |
CN111277444A (en) * | 2020-02-05 | 2020-06-12 | 苏州浪潮智能科技有限公司 | Switch fault early warning method and device |
CN111277444B (en) * | 2020-02-05 | 2022-12-27 | 苏州浪潮智能科技有限公司 | Switch fault early warning method and device |
CN111523647A (en) * | 2020-04-26 | 2020-08-11 | 南开大学 | Network model training method and device, and feature selection model, method and device |
CN111523647B (en) * | 2020-04-26 | 2023-11-14 | 南开大学 | Network model training method and device, feature selection model, method and device |
CN111860569A (en) * | 2020-06-01 | 2020-10-30 | 国网浙江省电力有限公司宁波供电公司 | Power equipment abnormity detection system and method based on artificial intelligence |
CN111708682A (en) * | 2020-06-17 | 2020-09-25 | 腾讯科技(深圳)有限公司 | Data prediction method, device, equipment and storage medium |
CN111708682B (en) * | 2020-06-17 | 2021-10-26 | 腾讯科技(深圳)有限公司 | Data prediction method, device, equipment and storage medium |
CN112395806A (en) * | 2020-11-11 | 2021-02-23 | 北京京航计算通讯研究所 | Method and device for predicting residual life of comprehensive transmission hydraulic system |
CN112395806B (en) * | 2020-11-11 | 2021-11-09 | 北京京航计算通讯研究所 | Method and device for predicting residual life of comprehensive transmission hydraulic system |
CN112418306A (en) * | 2020-11-20 | 2021-02-26 | 上海工业自动化仪表研究院有限公司 | Gas turbine compressor fault early warning method based on LSTM-SVM |
CN112418306B (en) * | 2020-11-20 | 2024-03-29 | 上海工业自动化仪表研究院有限公司 | Gas turbine compressor fault early warning method based on LSTM-SVM |
CN113076913A (en) * | 2021-04-16 | 2021-07-06 | 嘉兴毕格智能科技有限公司 | Aircraft engine fault prediction method |
CN113240099A (en) * | 2021-07-09 | 2021-08-10 | 北京博华信智科技股份有限公司 | LSTM-based rotating machine health state prediction method and device |
CN113657628A (en) * | 2021-08-20 | 2021-11-16 | 武汉霖汐科技有限公司 | Industrial equipment monitoring method and system, electronic equipment and storage medium |
CN113837477B (en) * | 2021-09-27 | 2023-06-27 | 西安交通大学 | Method, device and equipment for predicting power grid faults under typhoon disasters driven by data |
CN113837477A (en) * | 2021-09-27 | 2021-12-24 | 西安交通大学 | Data dual-drive power grid fault prediction method, device and equipment under typhoon disaster |
Also Published As
Publication number | Publication date |
---|---|
CN109814527B (en) | 2020-11-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109814527A (en) | Based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method and device | |
CN109472110B (en) | Method for predicting residual service life of aeroengine based on LSTM network and ARIMA model | |
CN109886430B (en) | Equipment health state assessment and prediction method based on industrial big data | |
Alaswad et al. | A review on condition-based maintenance optimization models for stochastically deteriorating system | |
CN109711453B (en) | Multivariable-based equipment dynamic health state evaluation method | |
CN110008079A (en) | Monitor control index method for detecting abnormality, model training method, device and equipment | |
CN110456756B (en) | Method suitable for online evaluation of global operation state in continuous production process | |
CN108520320A (en) | A kind of equipment life prediction technique based on multiple shot and long term memory network and Empirical Bayes | |
CN105095963A (en) | Method for accurately diagnosing and predicting fault of wind tunnel equipment | |
Liu et al. | Real-time quality monitoring and diagnosis for manufacturing process profiles based on deep belief networks | |
CN107037306A (en) | Transformer fault dynamic early-warning method based on HMM | |
CN106896219B (en) | The identification of transformer sub-health state and average remaining lifetime estimation method based on Gases Dissolved in Transformer Oil data | |
CN112257745B (en) | Hidden Markov-based coal mine underground system health degree prediction method and device | |
CN109884892A (en) | Process industry system prediction model based on crosscorrelation time lag grey correlation analysis | |
CN110121682A (en) | The method and system of omen subsequence in discovery time series | |
CN104200288A (en) | Equipment fault prediction method based on factor-event correlation recognition | |
KR20200017506A (en) | Device for diagnosing abnormalities in processes and how to diagnose abnormalities | |
CN112132394B (en) | Power plant circulating water pump predictive state evaluation method and system | |
CN108052092A (en) | A kind of subway electromechanical equipment abnormal state detection method based on big data analysis | |
CN111126489A (en) | Power transmission equipment state evaluation method based on ensemble learning | |
CN110757510A (en) | Method and system for predicting remaining life of robot | |
CN115526375A (en) | Space flight equipment predictive maintenance system | |
Tič et al. | Enhanced lubricant management to reduce costs and minimise environmental impact | |
CN108470699B (en) | intelligent control system of semiconductor manufacturing equipment and process | |
Le | Contribution to deterioration modeling and residual life estimation based on condition monitoring data |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |