CN112287596A - Engine residual life prediction method based on clustering and LSTM - Google Patents
Engine residual life prediction method based on clustering and LSTM Download PDFInfo
- Publication number
- CN112287596A CN112287596A CN202011000802.4A CN202011000802A CN112287596A CN 112287596 A CN112287596 A CN 112287596A CN 202011000802 A CN202011000802 A CN 202011000802A CN 112287596 A CN112287596 A CN 112287596A
- Authority
- CN
- China
- Prior art keywords
- data
- clustering
- prediction
- lstm
- engine
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 238000004458 analytical method Methods 0.000 claims abstract description 22
- 230000036541 health Effects 0.000 claims abstract description 18
- 230000035772 mutation Effects 0.000 claims abstract description 16
- 238000012544 monitoring process Methods 0.000 claims abstract description 6
- 238000004364 calculation method Methods 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 12
- 238000012935 Averaging Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 9
- 230000015556 catabolic process Effects 0.000 claims description 9
- 238000006731 degradation reaction Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 238000010586 diagram Methods 0.000 claims description 8
- 230000007774 longterm Effects 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 4
- 238000001514 detection method Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- 230000008034 disappearance Effects 0.000 claims description 3
- 230000000644 propagated effect Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000010006 flight Effects 0.000 abstract description 2
- 210000004027 cell Anatomy 0.000 description 30
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000004927 fusion Effects 0.000 description 4
- 238000012417 linear regression Methods 0.000 description 4
- 238000003064 k means clustering Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 238000007621 cluster analysis Methods 0.000 description 2
- 239000000306 component Substances 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008033 biological extinction Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 210000002364 input neuron Anatomy 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Geometry (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Operations Research (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Automation & Control Theory (AREA)
- Aviation & Aerospace Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
An engine residual life prediction method based on clustering and LSTM. The method belongs to the field of health management of aero-engines, and particularly relates to a clustering and LSTM-based engine residual life prediction method. The method for predicting the remaining service life of the engine based on the clustering and the LSTM is used for monitoring the health condition of the aircraft engine by combining the clustering and the LSTM and predicting the remaining service life of the engine so as to ensure the normal operation of flights. The prediction is carried out according to the following steps: 1) standardizing; 2) clustering analysis; 3) analyzing mutation; 4) predicting an LSTM model; 5) calculating a weight value; 6) multi-stage prediction. Experimental analysis shows that the ILSTMC method is effective when the residual service life of the aero-engine is predicted, and operators can better manage the health of the aero-engine by combining the prediction result of the scheme.
Description
Technical Field
The invention belongs to the field of health management of aero-engines, and particularly relates to a clustering and LSTM-based engine residual life prediction method.
Background
Aircraft engines are a core component of aircraft, and deployment is extremely important to the health management of the engine due to the complex variability between its internal components, coupled with long-term operation in a fairly intense operating environment. The prediction of the residual service life of the aircraft engine is of great significance for improving the safe and reliable work of the aircraft engine and guaranteeing the flight safety of the aircraft.
In the aspect of residual life prediction of the engine at present, a single-stage residual life prediction method is mostly used, the residual life prediction of the engine is not accurate, and the method cannot be well applied to health management of the engine.
Disclosure of Invention
Aiming at the problems, the invention provides an engine residual life prediction method based on clustering and LSTM, which utilizes the combination of clustering and LSTM to monitor the health condition of an aircraft engine and predict the residual life of the engine so as to ensure the normal operation of flights.
The technical scheme of the invention is as follows: the prediction is carried out according to the following steps:
1) and (3) standardization: inputting engine data, and carrying out sample normalization averaging processing;
2) clustering analysis: accurately obtaining the performance degradation data of the engine in multiple stages through clustering analysis on the basis of the step 1);
3) mutation analysis: on the basis of the step 2), predicating degradation data of multiple stages by adopting ILSTMC, and replacing common hidden layer nodes with memory blocks in the predicating process to ensure the storage of information and avoid the problems of long-term dependence and gradient disappearance;
4) prediction of the LSTM model: using the mutation points for prediction;
5) calculating a weight value: calculating the weight value of each door according to an algorithm formula of the LSTM;
6) multi-stage prediction: the data are related sequentially, firstly, the LSTM of the clustering treatment divided into a plurality of stages is carried out, and then prediction is carried out.
The step 1) is specifically as follows: firstly, normalizing and averaging input multidimensional aeroengine performance monitoring data to normalize the data to [0, 1] so as to avoid the possibility of positive and negative offset, and then averaging the normalized result, wherein the specific calculation formula of normalization is as follows:
wherein x isnormalFor normalized data, xtAs raw data, xmaxIs the maximum value, x, in the original data sampleminAt its minimum. The normalized averaged data was defined as the Overall Health Score (OHS) of the engine.
The step 2) is specifically as follows: the Sum of Squared Errors (SSE) in the clusters is used to find the optimal number k of clusters, and the number of clusters is determined by observing the inflection point diagram. The formula is as follows:
ω(i,j)=[X1+X2+…+Xk]-1 (4)
wherein, XkThe probability of each cluster is represented, omega represents the probability number of the cluster to which the sample belongs, m is a fuzzy coefficient and is generally 2, and mu is the central point of each cluster.
The step 3) is specifically as follows: and (4) utilizing clustering analysis to obtain clusters and catastrophe points, and dividing the data of the engine into a plurality of stages. The occurrence of the abrupt points may cause the remaining life condition of the engine to change, so the processing of the abrupt points becomes the basis of clustering. The detection of the mutation points is mainly to calculate the distance between each object and the central objects, and the specific calculation is shown in formula (2) to formula (4).
The error at the time t-1 can be calculated by the formula (11), the value of the activation function sigmoid of the forgetting gate at the mutation point, namely the value of sigma, is approximately 0, the input gate is approximately 1, and the cell at the previous time can not be transmitted by time and stored to the current time, so that the mutation point can be generated.
The step 4) is specifically as follows: using the discontinuities for prediction, the flow of information in the state of the hidden layer cell can be controlled by the value of the error term σ being in the input, forgetting and output gates of [0, 1 ]:
wherein, CtIs the long-term state of the cell at the present time, FtIs the value of the forgotten gate at the present time, Ct-1Is the cell state at time t-1, ItIs an input for the current time of day,is the instant state of the current moment, and the forgetting gate determines the previous moment Ct-1How much of the cell state of (2) remains until the present time Ct。
The LSTM model uses the ability to remove or add information to the cell state by carefully designing the structure of three gates, an input gate, a forgetting gate and an output gate, respectively, to protect and update the cell state. The input gate allows information to enter the memory cell, the forgetting gate allows cell information to be forgotten or removed from the input memory cell, and the output gate outputs information from the input memory cell.
The three gates cooperate closely to update the cell information. Firstly, effective information is extracted through tanh at an input gate, then, an activation function sigmoid is used for information screening, and the information updating degree of cells is determined, ItIs a scalar for measuring cell renewal, and the output of the output layer is htThe hidden layer will serve as the output of the LSTM.
An input gate:
forget the door:
an output gate:
where σ is the value of the activation function sigmoid, WiIs the weight value of the input gate, ht-1Is the hidden state value of the cell at time t-1, and b represents the bias term. WfIs the weight of the forgetting gate, WoThe weights of the output gates are increased, each weight gradient at the t moment in the prediction process can be calculated through the formula (12) to the formula (15) along with the increment of the data points in the prediction process, and then the weights are substituted into the formula (16) to the formula (19) to obtain the final weight gradient after summation, so that the weights are updated.
The specific calculation steps of step 5) are as follows:
5.1) reversely calculating the value of the error term sigma corresponding to each moment, wherein the reverse propagation of the error term is divided into two directions, namely the reverse propagation along the time is respectively calculated, namely the error term sigma-1 at the moment t-1 is calculated from the current moment t; the error term is propagated up one layer.
5.2) calculation of the gradient of the weights, the gradient of each weight, i.e. W, being calculated according to the corresponding error termfh,Wch,Woh,WihTheir gradient is the sum of the gradients at the respective time instants. It is therefore necessary to first determine their gradient at time t and then determine their final gradient.
Calculating an error term sigma at time to,t,σf,t,σi,t,σc,tFurther, a weight value W at time t is obtainedfh,Wch,Woh,WihNamely:
the final gradient is obtained by adding the gradients at the various times together:
the LSTM prediction process is used for directly predicting the data of the 1 st class obtained in the clustering process as the data of the 1 st stage of prediction; predicting the data of the 1 st and 2 nd classes as the data of the 2 nd order stage, and so on; and obtaining the results of the real value and the predicted value of the overall health score OHS of the engine.
The step 6) is specifically as follows: and (3) multi-stage residual life prediction, wherein data adopts a sequential association relation, firstly, clustering is carried out on the LSTM divided into a plurality of stages, and then prediction is carried out. In order to more accurately obtain the prediction effect, directly predicting the data of the 1 st class obtained by clustering as the data of the 1 st stage of prediction; then, iteration is carried out on the data for further prediction, and the clustered data of the 1 st and 2 nd classes are used as the data of the 2 nd stage for prediction; and predicting the data of the 1 st, 2 nd and 3 rd classes obtained by clustering as the data of the 3 rd stage, and then performing model comparison analysis.
The invention researches the prediction of the remaining life of the engine based on clustering and LSTM. The deep learning method can learn complex data characteristics, so that the multi-stage degradation characteristics of the engine are analyzed by combining clustering, the multi-stage remaining life of the engine is predicted, the method can be applied to health management of the engine to a certain extent, and the method has certain practical significance.
The method realizes the fusion of multi-dimensional data, establishes a prediction model based on the result of multi-stage information fusion, introduces real-time monitoring data by using the parameters of the multi-stage prediction model, adopts the algorithm of cluster analysis and LSTM fusion to realize the update and prediction of the parameters of the multi-stage model, finally predicts the performance degradation trend of the aero-engine and obtains the accurate prediction of the residual life of the aero-engine.
The invention has the following beneficial effects: the invention establishes a new prediction model according to the performance parameters of the aeroengine, the model integrates the advantages of an LSTM neural network and K-means clustering, the K-means clustering has the advantages of short time, high speed, simple thought, easy explanation and the like in the aspect of calculation, the LSTM can replace a common hidden layer node by using a memory block during prediction, the information storage can be ensured to span any delay, an error signal is returned to a time point before a long time, the network society is enabled to forget and be far away from a saturated state, and the problems of long-term dependence, gradient extinction, explosion and the like are avoided. Experimental analysis and verification are carried out through the algorithm, the calculated error of the ILSTMC in the residual life prediction is superior to that of other traditional models, experimental analysis shows that the ILSTMC method is effective in the residual life prediction of the aero-engine, so that an operator can better manage the health of the engine by combining the prediction result of the scheme, and the method has important significance in guaranteeing the flight safety of the airplane.
Drawings
FIG. 1 is a schematic structural framework of the present invention;
FIG. 2 is an OHS normalization chart for the 83 st engine;
FIG. 3 is a graph of cluster analysis results for the 83 th engine;
FIG. 4(a) is an ILSTMC stage 1 prediction diagram for the 83 st engine;
FIG. 4(b) is an ILSTMC stage 2 prediction diagram for the 83 st engine;
FIG. 4(c) is an ILSTMC stage 3 prediction diagram for the 83 st engine;
FIG. 5 is a comparison graph of RMSE of ILSTMC and other algorithm predictions for the 83 th engine;
fig. 6 is a multi-stage life cycle prediction error comparison diagram for the 83 th engine.
Detailed Description
In order to clearly explain the technical features of the present patent, the following detailed description of the present patent is provided in conjunction with the accompanying drawings.
As shown in figures 1-6, the clustering and LSTM-based method for predicting the residual life of the aircraft engine mainly comprises two modules, namely a clustering module and a prediction module
The clustering module comprises: the method mainly has the advantages that the K-means clustering is short in calculation time, high in speed, simple in thought, easy to explain and the like, input engine data are normalized to be between 0 and 1, and then the average value is obtained. Next, the Sum of Squared Errors (SSE) in the clusters is used to find the optimal number of clusters, and the number of clusters is determined by observing the inflection point diagram. Finally, clustering and catastrophe points are obtained by utilizing clustering analysis, and the data of the engine are divided into a plurality of stages. The occurrence of the abrupt points may cause the remaining life condition of the engine to change, so the processing of the abrupt points becomes the basis of clustering. The detection of the discontinuities is mainly a calculation of the distance of each object from these central objects.
The prediction module comprises: and performing multi-stage prediction and multi-stage prediction based on the mutation points obtained by clustering, wherein the data adopts a sequential association relationship, and firstly performing clustering treatment on the LSTM divided into a plurality of stages and secondly performing prediction. In order to more accurately obtain the prediction effect, directly predicting the data of the 1 st class obtained by clustering as the data of the 1 st stage of prediction; then, iteration is carried out on the data for further prediction, and the clustered data of the 1 st and 2 nd classes are used as the data of the 2 nd stage for prediction; and predicting the data of the 1 st, 2 nd and 3 rd classes obtained by clustering as the data of the 3 rd stage, and then performing model comparison analysis. Finally, the prediction error value of each engine is calculated, and the life cycle of the engine is divided in consideration of prediction of the remaining life of the engine, and the prediction errors in 5 stages of 60%, 70%, 80%, 90% and 95% are respectively extracted and compared with the errors of each method. And substituting the prediction result into a cost function to obtain RMSE, and comparing the RMSE with the prediction results of the traditional neural networks LSTM and RNN and the linear regression method LP.
As shown in fig. 1 to 5, the method for predicting the remaining life of the engine based on clustering and LSTM of the present invention comprises the following specific steps:
1) and (3) standardization: inputting engine data, and carrying out sample normalization averaging processing; firstly, multi-dimensional aeroengine performance monitoring data are input, normalization and averaging processing is carried out on the data, and the purpose is to improve the convergence speed and the calculation accuracy of the model.
Firstly, normalizing and averaging input multidimensional aeroengine performance monitoring data to normalize the data to [0, 1] so as to avoid the possibility of positive and negative offset, and then averaging the normalized result, wherein the specific calculation formula of normalization is as follows:
wherein x isnormalFor normalized data, xtAs raw data, xmaxIs the maximum value, x, in the original data sampleminAt its minimum. The normalized averaged data was defined as the Overall Health Score (OHS) of the engine.
2) Clustering analysis: accurately obtaining the performance degradation data of the engine in multiple stages through clustering analysis on the basis of the step 1);
and (4) clustering analysis, namely calculating the optimal number k of clusters by using Sum of Squared Errors (SSE) in the clusters, and judging the cluster number by observing an inflection point diagram of the cluster. The formula is as follows:
ω(i,j)=[X1+X2+…+Xk]-1 (4)
wherein, XkThe probability of each cluster is represented, omega represents the probability number of the cluster to which the sample belongs, m is a fuzzy coefficient and is generally 2, and mu is the central point of each cluster.
3) Mutation analysis: on the basis of the step 2), predicating degradation data of multiple stages by adopting ILSTMC, and replacing common hidden layer nodes with memory blocks in the predicating process to ensure the storage of information and avoid the problems of long-term dependence and gradient disappearance;
and (4) utilizing clustering analysis to obtain clusters and catastrophe points, and dividing the data of the engine into a plurality of stages. The occurrence of the abrupt points may cause the remaining life condition of the engine to change, so the processing of the abrupt points becomes the basis of clustering. The detection of the mutation points is mainly to calculate the distance between each object and the central objects, and the specific calculation is shown in formula (2) to formula (4).
The error at the time t-1 can be calculated by the formula (11), the value of the activation function sigmoid of the forgetting gate at the mutation point, namely the value of sigma, is approximately 0, the input gate is approximately 1, and the cell at the previous time can not be transmitted by time and stored to the current time, so that the mutation point can be generated.
4) Prediction of the LSTM model: using the discontinuities for prediction, the flow of information in the state of the hidden layer cell can be controlled by the value of the error term σ being in the input, forgetting and output gates of [0, 1 ]:
wherein, CtIs the long-term state of the cell at the present time, FtIs the value of the forgotten gate at the present time, Ct-1Is the cell state at time t-1, ItIs an input for the current time of day,is the instant state of the current moment, and the forgetting gate determines the previous moment Ct-1How much of the cell state of (2) remains until the present time Ct。
The LSTM model uses the ability to remove or add information to the cell state by carefully designing the structure of three gates, an input gate, a forgetting gate and an output gate, respectively, to protect and update the cell state. The input gate allows information to enter the memory cell, the forgetting gate allows cell information to be forgotten or removed from the input memory cell, and the output gate outputs information from the input memory cell.
The three gates cooperate closely to update the cell information. Firstly, effective information is extracted through tanh at an input gate, then, an activation function sigmoid is used for information screening, and the information updating degree of cells is determined, ItIs a scalar for measuring cell renewal, and the output of the output layer is htThe hidden layer will serve as the output of the LSTM.
An input gate:
forget the door:
an output gate:
where σ is the value of the activation function sigmoid, WiIs the weight value of the input gate, ht-1Is the hidden state value of the cell at time t-1, and b represents the bias term. WfIs the weight of the forgetting gate, WoThe weights of the output gates are increased, each weight gradient at the t moment in the prediction process can be calculated through the formula (12) to the formula (15) along with the increment of the data points in the prediction process, and then the weights are substituted into the formula (16) to the formula (19) to obtain the final weight gradient after summation, so that the weights are updated.
5) Calculating a weight value: calculating the weight value of each gate according to an algorithm formula of the LSTM, wherein the specific calculation steps are as follows:
5.1) reversely calculating the value of the error term sigma corresponding to each moment, wherein the reverse propagation of the error term is divided into two directions, namely the reverse propagation along the time is respectively calculated, namely the error term sigma-1 at the moment t-1 is calculated from the current moment t; the error term is propagated up one layer.
5.2) calculation of the gradient of the weights, the gradient of each weight, i.e. W, being calculated according to the corresponding error termfh,Wch,Woh,WihTheir gradient is the sum of the gradients at the respective time instants. It is therefore necessary to first determine their gradient at time t and then determine their final gradient.
Calculating an error term sigma at time to,t,σf,t,σi,t,σc,tFurther, a weight value W at time t is obtainedfh,Wch,Woh,WihNamely:
the final gradient is obtained by adding the gradients at the various times together:
the LSTM prediction process is used for directly predicting the data of the 1 st class obtained in the clustering process as the data of the 1 st stage of prediction; predicting the data of the 1 st and 2 nd classes as the data of the 2 nd order stage, and so on; and obtaining the results of the real value and the predicted value of the overall health score OHS of the engine.
6) Multi-stage prediction: and (3) multi-stage residual life prediction, wherein data adopts a sequential association relation, firstly, clustering is carried out on the LSTM divided into a plurality of stages, and then prediction is carried out. In order to more accurately obtain the prediction effect, directly predicting the data of the 1 st class obtained by clustering as the data of the 1 st stage of prediction; then, iteration is carried out on the data for further prediction, and the clustered data of the 1 st and 2 nd classes are used as the data of the 2 nd stage for prediction; and predicting the data of the 1 st, 2 nd and 3 rd classes obtained by clustering as the data of the 3 rd stage, and then performing model comparison analysis. And substituting the prediction result of the ILSTMC into a cost function to obtain RSME, and comparing the RSME with the prediction results of the traditional neural networks LSTM and RNN and a linear regression LP method.
The prediction error value of each engine is calculated, and the life cycle of the engine is divided in consideration of prediction of the remaining life of the engine, and the prediction errors in 5 stages of 60%, 70%, 80%, 90% and 95% are respectively extracted and compared with the errors of the respective methods.
And substituting the prediction result into a cost function to obtain RMSE, and comparing with errors of the prediction results of the traditional neural networks LSTM and RNN and the linear regression LP method.
Experiment of
The experiments were validated using the C-MAPSS dataset on the NASA website. The C-MAPSS data set has 4 subsets, each subset has 27-dimensional data, wherein the first 2 dimensions are engine numbers and the number of times of engine operation, the 3 dimensions correspond to the operating environment setting, the 22 dimensions correspond to the gas path performance data of the sensor, and each subset has different operating environment and fault number.
The gas circuit parameters reflect the health state of each gas circuit component or unit body, so the gas circuit performance parameters are key data influencing the overall performance condition of the engine and play a leading role in the prediction process. And 7 characterization parameters in the gas path performance parameters are used as decision factors of the overall health condition of the engine, and data fusion of multidimensional parameters is carried out to describe the performance degradation track of the engine.
Assuming that the sequence length of a sample is 5, and the input neuron at each moment is 1, sequentially bringing a training set and a test set of data into a neural network for training and testing; in the Adam algorithm of the LSTM, 1 sample value is taken out each time for batch processing, the learning rate η is set to be 0.01, the hidden layer size is 2 layers, the node number of each layer is 16 and 8 respectively, and 5 epoch periods are respectively taken by a training set and a test set to predict the remaining life of the engine.
And determining the clustering number k of each engine through calculation, wherein the k value also represents the specific stage number, then respectively carrying out model prediction on each engine, comparing the prediction results corresponding to each k value, and selecting the clustering number with the minimum error as the specific clustering number of each engine.
Taking the 83 th engine as an example, when using cluster and LSTM based ILSTMC model for prediction, the following comparison may illustrate the superiority of the cluster and LSTM based prediction method:
as shown in fig. 4, compared with RNN, LSTM, linear regression method, it was found that the overall health score OHS value of the engine is closer to the true value as the time point increases during the prediction process of the ILSTMC model;
in the case of multi-stage prediction, the RMSE of the last stage is reduced by 0.67% compared to LSTM, and on average each stage is reduced by 2.39% compared to LSTM;
as shown in fig. 5, for the full life cycle prediction, the error value is reduced by 0.58% from LSTM for 90% -95% life cycle, and the average error per stage life is reduced by 2.64% from LSTM.
While the invention has been described in terms of its preferred embodiments, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.
Claims (7)
1. A method for predicting the remaining life of an engine based on clustering and LSTM is characterized by comprising the following steps:
1) and (3) standardization: inputting engine data, and carrying out sample normalization averaging processing;
2) clustering analysis: accurately obtaining the performance degradation data of the engine in multiple stages through clustering analysis on the basis of the step 1);
3) mutation analysis: on the basis of the step 2), predicating degradation data of multiple stages by adopting ILSTMC, and replacing common hidden layer nodes with memory blocks in the predicating process to ensure the storage of information and avoid the problems of long-term dependence and gradient disappearance;
4) prediction of the LSTM model: using the mutation points for prediction;
5) calculating a weight value: calculating the weight value of each door according to an algorithm formula of the LSTM;
6) multi-stage prediction: the data are related sequentially, firstly, the LSTM of the clustering treatment divided into a plurality of stages is carried out, and then prediction is carried out.
2. The method for predicting the remaining life of the engine based on the clustering and the LSTM according to claim 1, wherein the step 1) is specifically as follows: firstly, normalizing and averaging input multidimensional aeroengine performance monitoring data to normalize the data to [0, 1] so as to avoid the possibility of positive and negative offset, and then averaging the normalized result, wherein the specific calculation formula of normalization is as follows:
wherein x isnormalFor normalized data, xtAs raw data, xmaxIs the maximum value, x, in the original data sampleminAt its minimum. The normalized averaged data was defined as the Overall Health Score (OHS) of the engine.
3. The method for predicting the remaining life of the engine based on the clustering and the LSTM according to claim 1, wherein the step 2) is specifically as follows: the Sum of Squared Errors (SSE) in the clusters is used to find the optimal number k of clusters, and the number of clusters is determined by observing the inflection point diagram. The formula is as follows:
ω(i,j)=[X1+X2+…+Xk]-1 (4)
wherein, XkThe probability of each cluster is represented, omega represents the probability number of the cluster to which the sample belongs, m is a fuzzy coefficient and is generally 2, and mu is the central point of each cluster.
4. The method for predicting the remaining life of the engine based on the clustering and the LSTM according to claim 1, wherein the step 3) is specifically as follows: and (4) utilizing clustering analysis to obtain clusters and catastrophe points, and dividing the data of the engine into a plurality of stages. The occurrence of the abrupt points may cause the remaining life condition of the engine to change, so the processing of the abrupt points becomes the basis of clustering. The detection of the mutation points is mainly to calculate the distance between each object and the central objects, and the specific calculation is shown in formula (2) to formula (4).
The error at the time t-1 can be calculated by the formula (11), the value of the activation function sigmoid of the forgetting gate at the mutation point, namely the value of sigma, is approximately 0, the input gate is approximately 1, and the cell at the previous time can not be transmitted by time and stored to the current time, so that the mutation point can be generated.
5. The method for predicting the remaining life of the engine based on the clustering and the LSTM according to claim 1, wherein the step 4) is specifically as follows: using the discontinuities for prediction, the flow of information in the state of the hidden layer cell can be controlled by the value of the error term σ being in the input, forgetting and output gates of [0, 1 ]:
wherein, CtIs the long-term state of the cell at the present time, FtIs the value of the forgotten gate at the present time, Ct-1Is the cell state at time t-1, ItIs an input for the current time of day,is the instant state of the current moment, and the forgetting gate determines the previous moment Ct-1How much of the cell state of (2) remains until the present time Ct。
The LSTM model uses the ability to remove or add information to the cell state by carefully designing the structure of three gates, an input gate, a forgetting gate and an output gate, respectively, to protect and update the cell state. The input gate allows information to enter the memory cell, the forgetting gate allows cell information to be forgotten or removed from the input memory cell, and the output gate outputs information from the input memory cell.
The three gates cooperate closely to update the cell information. Firstly, effective information is extracted through tanh at an input gate, then, an activation function sigmoid is used for information screening, and the information updating degree of cells is determined, ItIs a scalar for measuring cell renewal, and the output of the output layer is htThe hidden layer will serve as the output of the LSTM.
An input gate:
forget the door:
an output gate:
where σ is the value of the activation function sigmoid, WiIs the weight value of the input gate, ht-1Is the hidden state value of the cell at time t-1, and b represents the bias term. WfIs the weight of the forgetting gate, WoIs the weight of the output gate, increasing with the predicted process data pointEach weight gradient at the time t in the prediction process can be calculated through the formula (12) to the formula (15), and then the final weight gradient after summation is obtained by substituting the formula (16) to the formula (19), so as to update the weight.
6. The method for predicting the remaining life of the engine based on the clustering and the LSTM according to claim 1, wherein the specific calculation steps of the step 5) are as follows:
5.1) reversely calculating the value of the error term sigma corresponding to each moment, wherein the reverse propagation of the error term is divided into two directions, namely the reverse propagation along the time is respectively calculated, namely the error term sigma-1 at the moment t-1 is calculated from the current moment t; the error term is propagated up one layer.
5.2) calculation of the gradient of the weights, the gradient of each weight, i.e. W, being calculated according to the corresponding error termfh,Wch,Woh,WihTheir gradient is the sum of the gradients at the respective time instants. It is therefore necessary to first determine their gradient at time t and then determine their final gradient.
Calculating an error term sigma at time to,t,σf,t,σi,t,σc,tFurther, a weight value W at time t is obtainedfh,Wch,Woh,WihNamely:
the final gradient is obtained by adding the gradients at the various times together:
the LSTM prediction process is used for directly predicting the data of the 1 st class obtained in the clustering process as the data of the 1 st stage of prediction; predicting the data of the 1 st and 2 nd classes as the data of the 2 nd order stage, and so on; and obtaining the results of the real value and the predicted value of the overall health score OHS of the engine.
7. The method for predicting the remaining life of the engine based on the clustering and the LSTM according to claim 1, wherein the step 6) is specifically as follows: and (3) multi-stage residual life prediction, wherein data adopts a sequential association relation, firstly, clustering is carried out on the LSTM divided into a plurality of stages, and then prediction is carried out. In order to more accurately obtain the prediction effect, directly predicting the data of the 1 st class obtained by clustering as the data of the 1 st stage of prediction; then, iteration is carried out on the data for further prediction, and the clustered data of the 1 st and 2 nd classes are used as the data of the 2 nd stage for prediction; and predicting the data of the 1 st, 2 nd and 3 rd classes obtained by clustering as the data of the 3 rd stage, and then performing model comparison analysis.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011000802.4A CN112287596A (en) | 2020-09-22 | 2020-09-22 | Engine residual life prediction method based on clustering and LSTM |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011000802.4A CN112287596A (en) | 2020-09-22 | 2020-09-22 | Engine residual life prediction method based on clustering and LSTM |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112287596A true CN112287596A (en) | 2021-01-29 |
Family
ID=74422184
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011000802.4A Pending CN112287596A (en) | 2020-09-22 | 2020-09-22 | Engine residual life prediction method based on clustering and LSTM |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112287596A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113449472A (en) * | 2021-06-28 | 2021-09-28 | 哈尔滨工业大学 | Turbofan engine residual life prediction method |
CN113761722A (en) * | 2021-08-18 | 2021-12-07 | 航天科工海鹰集团有限公司 | Spacecraft multi-working-condition service life prediction method based on PCA |
CN114707234A (en) * | 2022-03-03 | 2022-07-05 | 电子科技大学 | Aero-engine residual life prediction method based on Bayesian deep learning |
CN117454671A (en) * | 2023-12-22 | 2024-01-26 | 广东力宏微电子有限公司 | Artificial intelligence-based field effect transistor life assessment method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104166787A (en) * | 2014-07-17 | 2014-11-26 | 南京航空航天大学 | Aero-engine remaining life prediction method based on multi-stage information fusion |
CN109033499A (en) * | 2018-06-07 | 2018-12-18 | 南京航空航天大学 | A kind of aero-engine method for predicting residual useful life of multistage consistency check |
CN109190301A (en) * | 2018-10-09 | 2019-01-11 | 南京工业大学 | Pivoting support high-precision life-span prediction method under a kind of variable working condition |
-
2020
- 2020-09-22 CN CN202011000802.4A patent/CN112287596A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104166787A (en) * | 2014-07-17 | 2014-11-26 | 南京航空航天大学 | Aero-engine remaining life prediction method based on multi-stage information fusion |
CN109033499A (en) * | 2018-06-07 | 2018-12-18 | 南京航空航天大学 | A kind of aero-engine method for predicting residual useful life of multistage consistency check |
CN109190301A (en) * | 2018-10-09 | 2019-01-11 | 南京工业大学 | Pivoting support high-precision life-span prediction method under a kind of variable working condition |
Non-Patent Citations (3)
Title |
---|
NING DING 等: "Journal bearing seizure degradation assessment and remaining useful life prediction based on long short-term memory neural network", 《MEASUREMENT》, 15 July 2020 (2020-07-15), pages 1 - 14 * |
刘君强: "基于超统计的多阶段航空发动机剩余寿命预测", 《北京航空航天大学学报》, 27 April 2020 (2020-04-27), pages 56 - 64 * |
战争热诚: "深度学习基础(6)——长短时记忆网络(LSTM)", 《HTTPS://WWW.CNBLOGS.COM/WJ-1314/ARTICLES/9921002.HTML》, 19 November 2018 (2018-11-19), pages 1 - 16 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113449472A (en) * | 2021-06-28 | 2021-09-28 | 哈尔滨工业大学 | Turbofan engine residual life prediction method |
CN113761722A (en) * | 2021-08-18 | 2021-12-07 | 航天科工海鹰集团有限公司 | Spacecraft multi-working-condition service life prediction method based on PCA |
CN114707234A (en) * | 2022-03-03 | 2022-07-05 | 电子科技大学 | Aero-engine residual life prediction method based on Bayesian deep learning |
CN114707234B (en) * | 2022-03-03 | 2023-04-25 | 电子科技大学 | Aero-engine residual life prediction method based on Bayesian deep learning |
CN117454671A (en) * | 2023-12-22 | 2024-01-26 | 广东力宏微电子有限公司 | Artificial intelligence-based field effect transistor life assessment method |
CN117454671B (en) * | 2023-12-22 | 2024-04-12 | 广东力宏微电子有限公司 | Artificial intelligence-based field effect transistor life assessment method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110321603B (en) | Depth calculation model for gas path fault diagnosis of aircraft engine | |
Zhao et al. | A recurrent neural network approach for remaining useful life prediction utilizing a novel trend features construction method | |
Zheng et al. | Long short-term memory network for remaining useful life estimation | |
CN112287596A (en) | Engine residual life prediction method based on clustering and LSTM | |
CN109766583B (en) | Aircraft engine life prediction method based on unlabeled, unbalanced and initial value uncertain data | |
CN112036084B (en) | Similar product life migration screening method and system | |
CN107786369B (en) | Power communication network security situation perception and prediction method based on IRT (intelligent resilient test) hierarchical analysis and LSTM (local Scale TM) | |
Li et al. | A light gradient boosting machine for remainning useful life estimation of aircraft engines | |
CN112734002B (en) | Service life prediction method based on data layer and model layer joint transfer learning | |
CN112257935B (en) | Aviation safety prediction method based on LSTM-RBF neural network model | |
CN114266278B (en) | Dual-attention network-based equipment residual service life prediction method | |
Liu et al. | Lithium-ion battery remaining useful life prediction with long short-term memory recurrent neural network | |
CN114218872B (en) | DBN-LSTM semi-supervised joint model-based residual service life prediction method | |
Wang et al. | A remaining useful life prediction model based on hybrid long-short sequences for engines | |
Jiang et al. | A fault diagnosis method for electric vehicle power lithium battery based on wavelet packet decomposition | |
Huang et al. | Bayesian neural network based method of remaining useful life prediction and uncertainty quantification for aircraft engine | |
Mo et al. | Few-shot RUL estimation based on model-agnostic meta-learning | |
Xiao et al. | A noise-boosted remaining useful life prediction method for rotating machines under different conditions | |
Hong et al. | Remaining useful life prediction using time-frequency feature and multiple recurrent neural networks | |
Xu et al. | High-accuracy health prediction of sensor systems using improved relevant vector-machine ensemble regression | |
CN115062528A (en) | Prediction method for industrial process time sequence data | |
Zhang et al. | Remaining useful life prediction for rolling bearings with a novel entropy-based health indicator and improved particle filter algorithm | |
Deng et al. | A remaining useful life prediction method with automatic feature extraction for aircraft engines | |
Jiang et al. | Multi-objective optimal placement of sensors based on quantitative evaluation of fault diagnosability | |
CN114580705B (en) | Method for predicting residual life of avionics product |
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 |