CN109814527B - Industrial equipment fault prediction method and device based on LSTM recurrent neural network - Google Patents
Industrial equipment fault prediction method and device based on LSTM recurrent neural network Download PDFInfo
- Publication number
- CN109814527B CN109814527B CN201910026193.0A CN201910026193A CN109814527B CN 109814527 B CN109814527 B CN 109814527B CN 201910026193 A CN201910026193 A CN 201910026193A CN 109814527 B CN109814527 B CN 109814527B
- Authority
- CN
- China
- Prior art keywords
- prediction
- feature
- vector
- fault
- 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.)
- Active
Links
Images
Landscapes
- Testing And Monitoring For Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method and a device for predicting the fault of industrial equipment based on an LSTM recurrent neural network, wherein the method comprises the following steps: acquiring state monitoring data sets of a plurality of sensors around target equipment, wherein the state monitoring data sets comprise monitoring data from 0 moment to the current moment; selecting a prediction characteristic containing preset fault information from the state monitoring data set by using a characteristic selection standard, wherein the characteristic selection standard comprises a correlation index and a monotonicity index; performing feature conversion on the predicted features to obtain predicted feature vectors; and performing single-step fault prediction, long-term fault prediction and residual life prediction on the target equipment according to the prediction feature vector and the fault prediction network model. The method can effectively avoid the insufficient prediction precision caused by unreasonable preset fault threshold, can give a confidence interval under the condition of single-step performance prediction, and can realize long-term prediction of the equipment performance and the residual life.
Description
Technical Field
The invention relates to the technical field of data-driven fault prediction, in particular to a fault prediction method and a fault prediction device for industrial equipment based on an LSTM (Long Short-Term Memory network) recurrent neural network.
Background
In the related art, the data-driven device fault prediction method is mainly implemented based on statistical analysis, bayesian Network, SVM (Support Vector Machine), HMM (Hidden Markov Model), NN (Neural Network), and the like. Although the method obtains good results under specific tasks and specific scenes, the method has the problems of low prediction precision, insufficient popularization capability and difficulty or incapability of long-term (long-term) prediction, and has difficulty in subsequent work such as operation and maintenance strategy optimization.
The main forms of fault prediction are residual life prediction and equipment performance prediction, known information is usually given in the form of sequence data (such as monitoring data sequences of industrial equipment), and the fault prediction is generally achieved by establishing a prediction model, and parameters of the prediction model can be learned from the sequence data. Therefore, the failure prediction problem can be regarded as a sequence learning problem.
In recent years, RNNs (Recurrent Neural Networks) have shown absolute advantages in solving sequence learning problems, such as precise timing problems, language models, simple pen drawing, voice recognition, and the like, and thus have attracted great attention in the field of fault prediction. The existing fault prediction method based on the RNN is generally implemented by two ways: the method takes the characteristics of T and T-T, T-2 T.time as input and the characteristics of T + T time as output, so as to achieve the purpose of performance prediction, however, the prediction precision is obviously reduced along with the increase of T, and when T is smaller, the capability of guiding maintenance decision is lost, so that the practical application is difficult; another method uses the RNN as a feature extraction model, and based on features input by the RNN, the remaining life is generally calculated by a preset fault threshold and an index model, so that advantages and characteristics of the RNN compared with other neural networks are not fully utilized.
Disclosure of Invention
The present invention is directed to solving, at least to some extent, one of the technical problems in the related art.
Therefore, one objective of the present invention is to provide a method for predicting a fault of an industrial device based on an LSTM recurrent neural network, which can effectively avoid the lack of prediction accuracy caused by unreasonable default fault thresholds, can give a confidence interval in the case of single-step performance prediction, and can realize long-term prediction of device performance and remaining life.
The invention also aims to provide a fault prediction device for industrial equipment based on the LSTM recurrent neural network.
In order to achieve the above object, an embodiment of the present invention provides a method for predicting a failure of an industrial device based on an LSTM recurrent neural network, including the following steps: step S101: acquiring a state monitoring data set of a plurality of sensors around target equipment, wherein the state monitoring data set comprises monitoring data from 0 moment to the current moment; step S102: selecting a prediction feature containing preset fault information from the state monitoring data set by using a feature selection standard, wherein the feature selection standard comprises a correlation index and a monotonicity index; step S103: performing feature conversion on the prediction features to obtain prediction feature vectors; step S104: and performing single-step fault prediction, long-term fault prediction and residual life prediction on the target equipment according to the prediction feature vector and the fault prediction network model.
The LSTM recurrent neural network-based industrial equipment fault prediction method can effectively avoid the insufficient prediction precision caused by unreasonable preset fault threshold values; the performance prediction value is obtained by sampling from Gaussian mixture distribution, the output is the parameters of the Gaussian mixture distribution, not only the current prediction value can be obtained, but also the distribution condition of the prediction value can be obtained, and a confidence interval can be given under the condition of single-step performance prediction; the long-term prediction of the equipment performance and the residual life can be realized without depending on machine learning methods such as a support vector machine and k-nearest neighbor.
In addition, the method for predicting the fault of the industrial equipment based on the LSTM recurrent neural network according to the embodiment of the invention can also have the following additional technical characteristics:
further, in one embodiment of the invention, the monitoring data includes one or more of temperature data, pressure data and deformability data.
Further, in one embodiment of the present invention, the feature selection criteria are:
Criteria＝α·Corr+(1-α)·Mon，
wherein alpha belongs to [0, 1] as a balance factor, Corr is the correlation index, and Mon is the monotonicity index;
wherein the ith characteristic sequence is f(i)The time span of the observation sequence is T, ft (i)For the sampling of the ith dimension feature at time t,
is mean value, df(i)Is f(i)The derivative of (c).
Further, in an embodiment of the present invention, the performing feature transformation on the predicted feature to obtain a predicted feature vector further includes: encoding a plurality of device states of the target device by using a one-hot vector to obtain a state identification vector; and obtaining a sensor monitoring vector according to the prediction characteristic, and obtaining the prediction characteristic vector according to the relative change of the sensor monitoring vector at different moments and the state identification vector.
Further, in an embodiment of the present invention, the method for training the fault prediction network model includes: acquiring monitoring historical data sets of a plurality of sensors around target equipment; processing the monitoring historical data set according to the step S102 and the step S103 to obtain a training feature vector; obtaining a training characteristic data set according to the monitoring historical data set, and building a training model based on a long-term and short-term memory unit and a Gaussian mixture model; and training the training model according to the training feature vector and the training feature data set to obtain the fault prediction network model.
In order to achieve the above object, another embodiment of the present invention provides an apparatus for predicting a failure of an industrial device based on an LSTM recurrent neural network, including: the system comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for acquiring a state monitoring data set of a plurality of sensors around target equipment, and the state monitoring data set comprises monitoring data from 0 moment to the current moment; the characteristic selection module is used for selecting a prediction characteristic containing preset fault information from the state monitoring data set by utilizing a characteristic selection standard, wherein the characteristic selection standard comprises a correlation index and a monotonicity index; the characteristic conversion module is used for carrying out characteristic conversion on the predicted characteristic to obtain a predicted characteristic vector; and the prediction module is used for performing single-step fault prediction, long-term fault prediction and residual life prediction on the target equipment according to the prediction feature vector and the fault prediction network model.
The fault prediction device based on the LSTM recurrent neural network industrial equipment can effectively avoid the insufficient prediction precision caused by unreasonable preset fault threshold; the performance prediction value is obtained by sampling from Gaussian mixture distribution, the output is the parameters of the Gaussian mixture distribution, not only the current prediction value can be obtained, but also the distribution condition of the prediction value can be obtained, and a confidence interval can be given under the condition of single-step performance prediction; the long-term prediction of the equipment performance and the residual life can be realized without depending on machine learning methods such as a support vector machine and k-nearest neighbor.
In addition, the fault prediction device based on the LSTM recurrent neural network industrial equipment according to the above embodiment of the present invention may further have the following additional technical features:
further, in one embodiment of the invention, the monitoring data includes one or more of temperature data, pressure data and deformability data.
Further, in one embodiment of the present invention, the feature selection criteria are:
Criteria＝α·Corr+(1-α)·Mon，
wherein alpha belongs to [0, 1] as a balance factor, Corr is the correlation index, and Mon is the monotonicity index;
wherein the ith characteristic sequence is f(i)The time span of the observation sequence is T, ft (i)For the sampling of the ith dimension feature at time t,
is mean value, df(i)Is f(i)The derivative of (c).
Further, in an embodiment of the present invention, the feature conversion module is further configured to encode the multiple device states of the target device by using a one-hot vector to obtain a state identification vector, obtain a sensor monitoring vector according to the predicted feature, and obtain the predicted feature vector according to the relative change of the sensor monitoring vector at different times and the state identification vector.
Further, in an embodiment of the present invention, the method further includes: the model training module is used for acquiring monitoring historical data sets of a plurality of sensors around target equipment, processing the monitoring historical data sets according to the feature selection module and the feature conversion module to obtain training feature vectors, obtaining training feature data sets according to the monitoring historical data sets, building a training model based on a long-short term memory unit and a Gaussian mixture model, and training the training model according to the training feature vectors and the training feature data sets to obtain the fault prediction network model.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a flow diagram of a method for fault prediction for industrial equipment based on an LSTM recurrent neural network in accordance with one embodiment of the present invention;
FIG. 2 is a flow diagram of a method for fault prediction for industrial equipment based on an LSTM recurrent neural network in accordance with an embodiment of the present invention;
FIG. 3 is a schematic diagram of an LSTM neuron according to one embodiment of the present invention;
FIG. 4 is a schematic diagram of a network model according to one embodiment of the invention;
fig. 5 is a schematic structural diagram of an apparatus for predicting a failure of an industrial device based on an LSTM recurrent neural network according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
The following describes a fault prediction method and device based on an LSTM recurrent neural network industrial device according to an embodiment of the present invention with reference to the drawings, and first, a fault prediction method based on an LSTM recurrent neural network industrial device according to an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a flow chart of a method for predicting a fault of an industrial device based on an LSTM recurrent neural network according to an embodiment of the present invention.
As shown in fig. 1, the LSTM recurrent neural network-based industrial equipment failure prediction method includes the following steps:
step S101: and acquiring a state monitoring data set of a plurality of sensors around the target equipment, wherein the state monitoring data set comprises monitoring data from 0 moment to the current moment. Wherein the monitoring data includes one or more of temperature data, pressure data, and deformability data.
It will be appreciated that, as shown in fig. 2, embodiments of the present invention first obtain historical data and real-time monitoring data of temperature, pressure, degree of deformation, etc. from pressure sensors, temperature sensors, other types of sensors, etc. deployed around the target component.
Step S102: and selecting a prediction characteristic containing preset fault information from the state monitoring data set by utilizing a characteristic selection standard, wherein the characteristic selection standard comprises a correlation index and a monotonicity index.
It is understood that the predicted features containing the preset fault information may be the features of the most fault information, as shown in fig. 2, the feature selection. The purpose of feature selection is to sort out the features that contain the most fault information. Since good features are generally consistent and monotonic with the degradation process of the device, embodiments of the present invention employ monotonicity indicators and consistency indicators for feature selection operations. The consistency index characterizes a linear relation between the characteristics and the equipment operation time, and the monotonicity index characterizes rising and falling trends of characteristic values.
Among them, in one embodiment of the present invention, the feature selection criteria are:
Criteria＝α·Corr+(1-α)·Mon，
wherein, alpha belongs to [0, 1] as a balance factor, Corr as a correlation index, and Mon as a monotonicity index;
wherein the ith characteristic sequence is f(i)The time span of the observation sequence is T, ft (i)For the sampling of the ith dimension feature at time t,
is mean value, df(i)Is f(i)The derivative of (c).
Specifically, note that the observed value sequence acquired by the ith sensor, i.e., the ith feature sequence is f(i)The time span of the observation sequence is T, ft (i)For the sampling of the ith dimension feature at time t,
is mean value, df(i)Is f(i)The derivative of (c) then has:
because the sensitivity of the characteristics to the faults is positively correlated with the two indexes, the linear combination of Corr and Mon can be used as the characteristic selection standard, and the alpha epsilon [0, 1] is recorded as a balance factor, and the following parameters are provided:
Criteria＝α·Corr+(1-α)·Mon。
step S103: and performing feature conversion on the predicted features to obtain predicted feature vectors.
Specifically, as shown in fig. 2, the features are transformed. Assuming that the dimension of the feature after the feature selection operation is k, and the feature after the conversion is S, the dimension of S is k +2, and the value S at the time tt＝(Δft，m1，m2) Wherein, Δ ft＝ft-ft-1，t∈[1，T]，f0＝f1，Δft∈Rk。(m1，m2) The vector is a one-hot vector and represents the running states of two devices: healthy and unhealthy. Specifically, m 11 means that the device is in a healthy state, and conversely, m 21 indicates that the equipment has a fault.
Step S104: and performing single-step fault prediction, long-term fault prediction and residual life prediction on the target equipment according to the prediction feature vector and the fault prediction network model.
It will be appreciated that the fault prediction is illustrated in figure 3. At any time t, based on the model parameters obtained by model training, single-step Prediction (One-stepPrediction), Long-term Prediction (Long-term Prediction) and residual Life Prediction (Remaining Useful Life Prediction) of the fault are carried out.
Specifically, the pseudo codes of the single-step fault prediction and the long-term fault prediction are shown in tables 1 and 2. t is tpThe steps of predicting the remaining life at the moment are as follows: continuously sampling from the current time
Up to
Time of day status indication vector
Then tpThe predicted value of the remaining life at the moment is as follows:
TABLE 1
TABLE 2
Further, in an embodiment of the present invention, a method for training a fault prediction network model includes: acquiring monitoring historical data sets of a plurality of sensors around target equipment; processing the monitoring historical data set according to the step S102 and the step S103 to obtain a training feature vector; obtaining a training characteristic data set according to the monitoring historical data set, and building a training model based on the long-short term memory unit and the Gaussian mixture model; and training the training model according to the training feature vector and the training feature data set to obtain a fault prediction network model.
It is understood that before the model training, the same processing is performed on the acquired data to acquire sensor history data, and then the feature selection is performed, the feature conversion is performed, and the prediction model after the model training is performed, as shown in fig. 2, by performing the steps S101 to S103. The specific process of model training comprises the following steps:
specifically, the network structure of the Model is constructed based on a Long Short Term Memory (LSTM) and a Gaussian Mixture Model (GMM) as shown in fig. 4.
Specifically, at each time point t, the LSTM network (neuron structure shown in fig. 3) features StAnd hidden state h of last momentt-1For input, the next time characteristic S is output via the full Connected Layert+1The variable of the probability distribution of (1) is assumed to be μ by using a Gaussian mixture model consisting of M normal distributionsiAnd σiThe mean value and the standard deviation of the ith Gaussian distribution respectively, and when the fault characteristic dimension is 1, the following steps are carried out:
wherein the content of the first and second substances,
in addition, the present example employs a classification Distribution (p)1，p2) Vector (m) indicating the operating state of a device1，m2) Modeling is performed to satisfy p1+p 21. Through the combination of the LSTM and the GMM, the model provided by the technology not only can predict the characteristic value of the subsequent time, but also can provide the distribution of the predicted value of the subsequent time.
Setting the initial hidden state and the input characteristic as h respectively0＝0，S0＝(0，1，0)，ytThe prediction model can be split into M +1 parts, which respectively correspond to M normal distributions and 1 class distribution, and in this case, the prediction model can be expressed as:
ht＝LSTM(St-1，ht-1)
yt＝Wyht+by
because the normal distribution standard deviation has nonnegativity, the technology utilizes the exp operator and the softmax operator to perform the following operations on the probability distribution parameters:
the embodiment of the invention carries out model training by minimizing reconstruction error (Reconstructionerror), the reconstruction error can be obtained by output calculation of a training set and a full-link layer, and a loss function LRAs follows:
Loss＝LR＝Ls+Lp
the method for predicting the fault of the industrial equipment based on the LSTM recurrent neural network is further explained by the specific embodiment.
Further, in an embodiment of the present invention, performing feature transformation on the predicted features to obtain a predicted feature vector, further includes: coding a plurality of equipment states of the target equipment by utilizing the one-hot vector to obtain a state identification vector; and obtaining a sensor monitoring vector according to the predicted characteristic, and obtaining a predicted characteristic vector according to the relative change of the sensor monitoring vector at different moments and the state identification vector.
Specifically, as shown in fig. 2, the fault prediction method according to the embodiment of the present invention specifically includes:
an offline stage: acquiring a sensor state monitoring historical data set of target equipment; by usingThe correlation index (Corr metric) and the monotonicity index (Mon metric) select k sensor monitor values containing the most fault information as features, and a sensor monitor vector f ═ f1，f2，...，fk](ii) a Coding the states of the devices by using one-hot vectors to obtain a state identification vector m ═ m1，m2，...，ml](ii) a Combining the relative change of the sensor monitoring vector at different moments with the state identification vector to form a final feature vector, wherein the feature vector is as follows by taking the moment t as an example: st＝[Δft，m]Wherein, Δ ft＝ft-ft-1(ii) a Deriving training feature data sets from historical data
Wherein the content of the first and second substances,
constructing a fault prediction network model M shown in FIG. 3, wherein the structure of the LSTM neuron is shown in FIG. 2; and training M by using an end-to-end method by taking S as input.
An online stage: the real-time equipment sensor monitoring data is processed in the same way as the offline stage, and a characteristic vector S from 0 moment to the current moment (taking t moment as an example) is obtained1：t＝{s1，s2，...，st}; performing one-step performance prediction and long-term performance prediction according to the procedures shown in the table 1 and the table 2 respectively; and calculating the residual life of the equipment according to the long-term performance evaluation result and the current moment.
In summary, the embodiment of the invention has the following advantages:
1) the failure threshold need not be preset. According to the method provided by the embodiment of the invention, the discretization working condition of the industrial equipment is partially input in the form of the one-hot vector, so that the downtime of the equipment is learned and output in the form of the one-hot vector, and therefore, a fault threshold does not need to be preset, and the insufficient prediction precision caused by unreasonable preset of the fault threshold can be effectively avoided.
2) In the case of single-step performance prediction, a confidence interval can be given. Different from the method of directly outputting the predicted value of the equipment performance or the predicted value of the residual life, the method of the embodiment of the invention obtains the predicted value of the performance by sampling from the Gaussian mixture distribution, and outputs the parameters of the Gaussian mixture distribution, so that the method not only can obtain the current predicted value, but also can obtain the distribution situation of the predicted value.
3) Long-term (long-term) performance prediction is enabled. The method provided by the embodiment of the invention does not depend on machine learning methods such as a Support Vector Machine (SVM), k-nearest neighbor (KNN) and the like, and can realize long-term prediction (long-term prediction) on the performance and the residual life of the equipment.
According to the fault prediction method based on the LSTM recurrent neural network industrial equipment, which is provided by the embodiment of the invention, the insufficient prediction precision caused by unreasonable preset fault threshold can be effectively avoided; the performance prediction value is obtained by sampling from Gaussian mixture distribution, the output is the parameters of the Gaussian mixture distribution, not only the current prediction value can be obtained, but also the distribution condition of the prediction value can be obtained, and a confidence interval can be given under the condition of single-step performance prediction; the long-term prediction of the equipment performance and the residual life can be realized without depending on machine learning methods such as a support vector machine and k-nearest neighbor.
Next, a failure prediction apparatus for an industrial device based on an LSTM recurrent neural network according to an embodiment of the present invention will be described with reference to the drawings.
Fig. 5 is a schematic structural diagram of an apparatus for predicting a fault of an industrial device based on an LSTM recurrent neural network according to an embodiment of the present invention.
As shown in fig. 5, the LSTM recurrent neural network-based industrial equipment failure prediction apparatus 10 includes: a data acquisition module 100, a feature selection module 200, a feature conversion module 300, and a prediction module 400.
The data acquisition module 100 is configured to acquire a state monitoring data set of a plurality of sensors around a target device, where the state monitoring data set includes monitoring data from time 0 to a current time. The feature selection module 200 is configured to select a predictive feature including preset fault information from the condition monitoring dataset by using a feature selection criterion, where the feature selection criterion includes a relevance index and a monotonicity index. The feature transformation module 300 is configured to perform feature transformation on the predicted features to obtain predicted feature vectors. The prediction module 400 is used for performing single-step fault prediction, long-term fault prediction and residual life prediction on the target device according to the prediction feature vector and the fault prediction network model. The device 10 of the embodiment of the invention can effectively avoid the insufficient prediction precision caused by unreasonable preset fault threshold, can give a confidence interval in the single-step performance prediction, and can realize the long-term prediction of the equipment performance and the residual life.
Further, in one embodiment of the invention, the monitoring data includes one or more of temperature data, pressure data, and deformability data.
Further, in one embodiment of the present invention, the feature selection criteria are:
Criteria＝α·Corr+(1-α)·Mon，
wherein, alpha belongs to [0, 1] as a balance factor, Corr as a correlation index, and Mon as a monotonicity index;
wherein the ith characteristic sequence is f(i)The time span of the observation sequence is T, ft (i)For the sampling of the ith dimension feature at time t,
is mean value, df(i)Is f(i)The derivative of (c).
Further, in an embodiment of the present invention, the feature conversion module is further configured to encode the multiple device states of the target device by using a one-hot vector to obtain a state identification vector, obtain a sensor monitoring vector according to the predicted feature, and obtain a predicted feature vector according to the relative change of the sensor monitoring vector at different times and the state identification vector.
Further, in an embodiment of the present invention, the apparatus of the embodiment of the present invention further includes: and a model training module. The model training module is used for acquiring monitoring historical data sets of a plurality of sensors around the target equipment, processing the monitoring historical data sets according to the feature selection module 200 and the feature conversion module 300 to obtain training feature vectors, acquiring training feature data sets according to the monitoring historical data sets, building a training model based on the long-short term memory unit and the Gaussian mixture model, and training the training model according to the training feature vectors and the training feature data sets to obtain a fault prediction network model.
It should be noted that the foregoing explanation of the embodiment of the method for predicting the fault of the industrial device based on the LSTM recurrent neural network is also applicable to the device for predicting the fault of the industrial device based on the LSTM recurrent neural network of the embodiment, and details are not repeated here.
According to the fault prediction device based on the LSTM recurrent neural network industrial equipment, which is provided by the embodiment of the invention, the defect of insufficient prediction precision caused by unreasonable preset fault threshold can be effectively avoided; the performance prediction value is obtained by sampling from Gaussian mixture distribution, the output is the parameters of the Gaussian mixture distribution, not only the current prediction value can be obtained, but also the distribution condition of the prediction value can be obtained, and a confidence interval can be given under the condition of single-step performance prediction; the long-term prediction of the equipment performance and the residual life can be realized without depending on machine learning methods such as a support vector machine and k-nearest neighbor.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
In the present invention, unless otherwise expressly stated or limited, the first feature "on" or "under" the second feature may be directly contacting the first and second features or indirectly contacting the first and second features through an intermediate. Also, a first feature "on," "over," and "above" a second feature may be directly or diagonally above the second feature, or may simply indicate that the first feature is at a higher level than the second feature. A first feature being "under," "below," and "beneath" a second feature may be directly under or obliquely under the first feature, or may simply mean that the first feature is at a lesser elevation than the second feature.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (8)
1. An LSTM recurrent neural network-based industrial equipment fault prediction method is characterized by comprising the following steps:
step S101: acquiring a state monitoring data set of a plurality of sensors around target equipment, wherein the state monitoring data set comprises monitoring data from 0 moment to the current moment;
step S102: selecting criteria from the shape using featuresSelecting a prediction characteristic containing preset fault information in the state monitoring data set, wherein the characteristic selection standard comprises a correlation index and a monotonicity index, and the characteristic selection standard is as follows: criterion ═ α · Corr + (1- α) · Mon, where α ∈ [0,1 ∈ Mon]Is an equilibrium factor, Corr is the correlation index, and Mon is the monotonicity index;
wherein the ith characteristic sequence is f(i)The time span of the observation sequence is T,
for the sampling of the ith dimension feature at time t,
is mean value, df(i)Is f(i)A derivative of (a);
step S103: performing feature conversion on the prediction features to obtain prediction feature vectors;
step S104: and performing single-step fault prediction, long-term fault prediction and residual life prediction on the target equipment according to the prediction feature vector and the fault prediction network model.
2. The LSTM recurrent neural network-based industrial device fault prediction method of claim 1, wherein the monitored data includes one or more of temperature data, pressure data, and deformability data.
3. The LSTM recurrent neural network-based industrial device failure prediction method of claim 1, wherein said feature transforming said predicted features to obtain a predicted feature vector further comprises:
encoding a plurality of device states of the target device by using a one-hot vector to obtain a state identification vector;
and obtaining a sensor monitoring vector according to the prediction characteristic, and obtaining the prediction characteristic vector according to the relative change of the sensor monitoring vector at different moments and the state identification vector.
4. The LSTM recurrent neural network-based industrial device fault prediction method of any of claims 1-3, wherein the training method of the fault prediction network model comprises:
acquiring monitoring historical data sets of a plurality of sensors around target equipment;
processing the monitoring historical data set according to the step S102 and the step S103 to obtain a training feature vector;
obtaining a training characteristic data set according to the monitoring historical data set, and building a training model based on a long-term and short-term memory unit and a Gaussian mixture model;
and training the training model according to the training feature vector and the training feature data set to obtain the fault prediction network model.
5. An industrial equipment fault prediction device based on an LSTM recurrent neural network is characterized by comprising:
the system comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for acquiring a state monitoring data set of a plurality of sensors around target equipment, and the state monitoring data set comprises monitoring data from 0 moment to the current moment;
a feature selection module, configured to select a prediction feature including preset fault information from the state monitoring data set by using a feature selection criterion, where the feature selection criterion includes a relevance index and a monotonicity index, and the feature selection criterion is: criterion ═ α · Corr + (1- α) · Mon, where α ∈ [0,1 ∈ Mon]Is an equilibrium factor, Corr is the correlation index, and Mon is the monotonicity index;
wherein the ith characteristic sequence is f(i)The time span of the observation sequence is T,
for the sampling of the ith dimension feature at time t,
is mean value, df(i)Is f(i)A derivative of (a);
the characteristic conversion module is used for carrying out characteristic conversion on the predicted characteristic to obtain a predicted characteristic vector;
and the prediction module is used for performing single-step fault prediction, long-term fault prediction and residual life prediction on the target equipment according to the prediction feature vector and the fault prediction network model.
6. The LSTM recurrent neural network-based industrial device failure prediction apparatus of claim 5, wherein the monitored data includes one or more of temperature data, pressure data, and deformability data.
7. The LSTM recurrent neural network-based industrial device fault prediction apparatus of claim 5, wherein the feature transformation module is further configured to encode a plurality of device states of the target device by using a one-hot vector to obtain a state identification vector, obtain a sensor monitoring vector according to the prediction feature, and obtain the prediction feature vector according to relative changes of the sensor monitoring vector at different times and the state identification vector.
8. The LSTM recurrent neural network-based industrial device fault prediction apparatus of any of claims 5-7, further comprising:
the model training module is used for acquiring monitoring historical data sets of a plurality of sensors around target equipment, processing the monitoring historical data sets according to the feature selection module and the feature conversion module to obtain training feature vectors, obtaining training feature data sets according to the monitoring historical data sets, building a training model based on a long-short term memory unit and a Gaussian mixture model, and training the training model according to the training feature vectors and the training feature data sets to obtain the fault prediction 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 CN109814527A (en) | 2019-05-28 |
| CN109814527B true 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) |
Families Citing this family (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287583B (en) * | 2019-06-21 | 2021-01-29 | 上海交通大学 | Industrial equipment residual life prediction method based on cyclic 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 |
| CN110764065B (en) * | 2019-10-16 | 2021-10-08 | 清华大学 | Radar fault diagnosis method based on time sequence reconstruction |
| 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 |
| 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 |
| 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 |
| 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 |
| 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 |
| 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 |
| 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 |
| 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 |
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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109814527A (en) | 2019-05-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109814527B (en) | Industrial equipment fault prediction method and device based on LSTM recurrent neural network | |
| CN109472110B (en) | Method for predicting residual service life of aeroengine based on LSTM network and ARIMA model | |
| CN111222549B (en) | Unmanned aerial vehicle fault prediction method based on deep neural network | |
| Aydin et al. | Using LSTM networks to predict engine condition on large scale data processing framework | |
| CN109711453B (en) | Multivariable-based equipment dynamic health state evaluation method | |
| CN112527788A (en) | Method and device for detecting and cleaning abnormal value of transformer monitoring data | |
| JP2019527413A (en) | Computer system and method for performing root cause analysis to build a predictive model of rare event occurrences in plant-wide operations | |
| CN111220387B (en) | Vehicle bearing residual life prediction method based on multi-feature-quantity correlation vector machine | |
| CN108664009B (en) | Stage division and fault detection method based on correlation analysis | |
| CN105468850B (en) | Electronic product degradation trend prediction technique based on more residual error regression forecasting algorithms | |
| CN110309537B (en) | Intelligent health prediction method and system for aircraft | |
| CN107037306A (en) | Transformer fault dynamic early-warning method based on HMM | |
| CN111340282A (en) | DA-TCN-based method and system for estimating residual service life of equipment | |
| KR102005138B1 (en) | Device abnormality presensing method and system using thereof | |
| CN106896219A (en) | The identification of transformer sub-health state and average remaining lifetime method of estimation based on Gases Dissolved in Transformer Oil data | |
| Le et al. | Multi-branch Hidden semi-Markov modeling for RUL prognosis | |
| CN113052302B (en) | Machine health monitoring method and device based on cyclic neural network and terminal equipment | |
| Liao et al. | Nonparametric and semi-parametric sensor recovery in multichannel condition monitoring systems | |
| Wang et al. | Three‐stage feature selection approach for deep learning‐based RUL prediction methods | |
| Xie et al. | The application of echo state network and recurrent multilayer perceptron in rotating machinery fault prognosis | |
| CN116677570A (en) | Fault early warning method and system based on cabin temperature monitoring of offshore wind turbine | |
| Atamuradov et al. | Segmentation based feature evaluation and fusion for prognostics | |
| CN113849479A (en) | Comprehensive energy supply station oil tank leakage detection method based on instant learning and self-adaptive threshold | |
| Liu et al. | On-line adaptive data-driven fault prognostics of complex systems | |
| Yan et al. | Remaining Useful Life Interval Prediction for Complex System Based on BiGRU Optimized by Log-Norm |
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 |