CN113722833B - Turbofan engine residual service life prediction method based on double-channel long-short-term memory network - Google Patents
Turbofan engine residual service life prediction method based on double-channel long-short-term memory network Download PDFInfo
- Publication number
- CN113722833B CN113722833B CN202111053571.8A CN202111053571A CN113722833B CN 113722833 B CN113722833 B CN 113722833B CN 202111053571 A CN202111053571 A CN 202111053571A CN 113722833 B CN113722833 B CN 113722833B
- Authority
- CN
- China
- Prior art keywords
- service life
- short
- time
- memory network
- residual service
- 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
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/04—Ageing analysis or optimisation against ageing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Abstract
The method comprises the steps of firstly measuring the variability of sensor data indexes for monitoring the state of an engine, obtaining the sensor data indexes with variability, then utilizing the two-channel long-short-time memory neural network to process the difference value between the data indexes, designing a convolution neural network module to extract local time characteristics of a long-short-time memory neural network sequence output result, then using the local time characteristics for the input of a two-layer fully-connected neural network to predict the residual service life of the engine, and finally using a predicted value at the last moment as a buffer for the predicted result at the current moment to smoothly calibrate the current predicted value. The method effectively reduces the interference of fault noise, improves the capability of long-time memory neural network processing time sequences, and finally improves the accuracy of fitting the residual service life.
Description
Technical Field
The invention belongs to the field of life prediction in equipment health management, and particularly relates to a method for predicting the residual service life of a turbine engine based on a double-channel long-short-time memory network.
Background
The turbine engine is the heart of the aircraft, powering the flight of the aircraft. However, since the engine is often operated in a high temperature and high pressure environment, the problem of failure is difficult to avoid. If the engine is light, the aircraft cannot take off, passengers change the labels, the reputation of the airlines is damaged, and the like, and if the engine is heavy, the aircraft is damaged, and the personnel are injured. It is therefore of great importance how to accurately predict the remaining service life of an engine before a fault occurs. At present, the traditional deep learning method achieves good effect, but still faces the following problems:
(1) The influence of the time characteristics on life prediction of the turbine engine is changed when the turbine engine operates in different environments, and for this phenomenon, how to select useful time characteristics is a problem to be solved, so as to avoid the occurrence of redundancy or invalid characteristics.
(2) For the time feature, researchers often let the model pay attention to the time feature at a certain moment, neglect the difference value of the time features at two different moments, for example, if the time feature value is always bigger, but the feature difference value is smaller, how to use the two parameters to predict the service life, reduce the noise influence, and make the model more robust is also a problem to be solved.
(3) In general, the service life of the turbine engine is smooth and stable, that is, the length of the residual service life of the engine in a certain period of time is not quite different, the number of times of fluctuation is relatively small, but under the working of a severe environment, the data of a sensor return system are often unclean, the traditional deep learning method learns and predicts according to the data, the residual service life predicted by using a neural network can fluctuate up and down, and the residual service life curve shows a saw-tooth shape, which causes larger deviation from the actual residual service life.
Disclosure of Invention
In order to solve the problems, a method for predicting the residual service life of a turbofan engine with a double-channel long-short-term memory network is provided. Firstly, a self-adaptive characteristic selection method is adopted, characteristics with predictability are selected by utilizing characteristic variability (prognosity) according to different data sets, then a double-channel long-short-time memory network is utilized to process time characteristic values and characteristic difference values, a convolutional neural network is utilized to extract output characteristics of each time after long-short-time memory network processing, then the residual service life is predicted through a full-connection neural network, finally, a momentum smoothing method is provided for processing an actual residual service life curve according to the problem that a service life curve is saw-tooth-shaped under the inspired of gradient momentum.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
a service life prediction method based on a dual-channel long-short-time memory network comprises the following steps:
A. preprocessing data, selecting characteristics useful for model prediction, normalizing the data, accelerating the convergence of the model, and expanding the data in three aspects of characteristic selection, normalization processing and time window processing;
a. solving the prognosity of the features, and selecting the features with larger change in a certain time range for training the model, wherein the prognosity formula is as follows:
x j a measurement vector representing a feature on the jth system, the variable M being the number of monitored systems, N j Is the number of measurements on the j-th system, it can be observed that for some features, if prognosity is equal to 0 or NaN, these features are removed, forming a new sample set;
b. data normalization was performed using z-score:
u represents the mean value of all selected features, σ represents the standard deviation of all selected features, and x represents a certain selected feature value;
c. dividing a time window, the window width being denoted as N t The sliding stride is denoted s, the first time window input1= [ x ] 1 ,x 2 ,...,x Nf ],x i A certain time feature vector is represented and,representing a temporal feature vector x i At the first value of the time window, and so on;
a second time window (time feature difference) Input 2 =[d 1 ,d 2 ,...,d Nf ],d i Representing the difference of a certain time feature vector;
B. dual channel long and short term memory network processing Input 1 And Input 2 Then, two Output outputs are obtained 1 =[h 1 ,h 2 ,...,h hidden_size ]And Output 2 =[g 1 ,g 2 ,...,g hidden_size ],h i Representing long and short term memory network processing Input 1 Vector g i Representing long and short term memory network processing Input 2 The vector of the following;
will Output 1 After (N) t -1) line vectors and Output 2 Direct addition, output is obtained, output= [ o ] 1 ,o 2 ,...,o hidden_size ],o i From h i And g i Adding to obtain;
C. the service life prediction is divided into two parts, wherein one part is to extract local time characteristics of the two-channel long-short-time memory network sequence output by using a convolutional neural network, and the other part is to predict the residual service life by using a fully-connected neural network;
D. considering the buffer relation of the residual service life of the turbine engine in the earlier stage to the residual service life of the current moment, inspiring the reduction of the momentum gradient, adopting a momentum smoothing residual service life method for a test set, and adopting the following formula:
predict t =k×y t +(1-k)×predict t-1 ,0≤k≤1 (8)
y t the residual service life at the time t is predicted by using a double-channel long-short-time memory network t-1 Is the prediction result after the previous moment is smoothed, and predicts t Is the remaining service life after the current t moment is smoothed, k represents y t In the prediction t The ratio of (3) is calculated.
The beneficial effects of the invention are as follows:
the utility model has the advantages of put forward a binary channels long and short time memory network model, it can learn the moment value and the moment difference of time characteristic, then through convolutional neural network processing, predict remaining life, finally through momentum smoothing's effect, predicted remaining life can more laminate true remaining life curve, binary channels long and short time memory network has improved the learning ability of network owing to having learned the information of two dimensions of time characteristic, adopt L2 regularization, dropout technique and verification to stop the technique soon to prevent the model from fitting excessively simultaneously.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a time window diagram for data preprocessing;
FIG. 3 is a diagram of a dual channel long and short time memory network;
FIG. 4 is a diagram of a residual life prediction block;
FIG. 5 is a graph of the remaining useful life of the FD001 sub-dataset portion engine unit;
fig. 6 is a momentum smoothing contrast diagram.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
Referring to fig. 1, a method for predicting residual service life based on a dual-channel long-short-time memory network comprises the following steps:
A. preprocessing data, selecting characteristics useful for model prediction, normalizing the data, accelerating the convergence of the model, and expanding the data in three aspects of characteristic selection, normalization processing and time window processing;
a. table 1 lists 21 sensor signature data for each engine monitored;
table 1 sensor introduction
Solving the prognosity of the monitoring data, and selecting out the characteristics with larger change for training the model, wherein the prognosity formula is as follows:
xj represents the measurement vector of a certain feature on the jth system, the variable M is the number of monitored systems, nj is the number of measurements on the jth system, and for certain features, if their prognosity is equal to 0 or NaN, these features are removed to form a new sample set;
b. data normalization was performed using z-score:
u represents the mean value of all selected features, σ represents the standard deviation of all selected features, and x represents a certain selected feature value;
c. referring to FIG. 2, time windows are divided, with window width denoted N t The sliding stride is denoted s, the first time window Input 1 =[x 1 ,x 2 ,...,x Nf ],x i A certain time feature vector is represented and,representing a temporal feature vector x i At the first value of the time window, and so on;
a second time window (time feature difference) Input 2 =[d 1 ,d 2 ,...,d Nf ],d i Representing the difference of a certain time feature vector;
B. referring to FIG. 3, a dual channel long and short time memory network processes Input 1 And Input 2 After that, two outputs Output can be obtained 1 =[h 1 ,h 2 ,...,h hidden_size ]And Output 2 =[g 1 ,g 2 ,...,g hidden_size ],h i Representing long and short term memory network processing Input 1 Vector g i Representing long and short term memory network processing Input 2 The vector of the following;
will Output 1 After (N) t -1) line vectors and Output 2 Direct addition, output is obtained, output= [ o ] 1 ,o 2 ,...,o hidden_size ],o i From h i And g i Adding to obtain;
C. referring to fig. 4, the life prediction may be divided into two parts, one part is to extract local time characteristics of the two-channel long-short-time memory network sequence output by using the convolutional neural network, and the other part is to predict the remaining service life by using the fully connected neural network;
D. considering the buffer relation of the residual service life of the turbine engine in the earlier stage to the residual service life of the current moment, inspiring the reduction of the momentum gradient, adopting a method for smoothing the residual service life by momentum for a test set, and adopting the following formula:
predict t =k×y t +(1-k)×predict t-1 ,0≤k≤1 (16)
y t the residual service life at the time t is predicted by using a double-channel long-short-time memory network t-1 Is the prediction result after the previous moment is smoothed, and predicts t Is the remaining service life after the current t moment is smoothed, k represents y t In the prediction t The proportion of the components;
E. to demonstrate the effectiveness of the method of the present invention, it is compared to different methods, further described, commercial modular aviation propulsion system simulation (C-MAPSS) raw data sets are detailed in table 2:
table 2 dataset introduction
The method comprises the steps of preprocessing data, processing time characteristics and characteristic difference values by utilizing a double-channel long-short time memory network, and finally smoothly predicting a residual service life curve. The experimental results of the method of the invention and some advanced methods are compared and analyzed, and are shown in Table 3 in detail:
TABLE 3 comparison of effects of the methods
The score formula and RMSE formula are as follows:
predict i representing the predicted value, RUL i Indicating the true remaining service life, N indicating the number of all sample data, a graph of the remaining service life of a part of the engine units is shown in fig. 5, table 4 describes the effect of using momentum smoothing, and fig. 6 compares the remaining service life curves after using momentum smoothing.
Table 4 k value influence effect
Claims (3)
1. The method for predicting the residual service life of the turbofan engine based on the double-channel long-short-time memory network is characterized by comprising the following steps of:
A. preprocessing the data, and expanding the data in three aspects of feature selection, standardization processing and time window processing;
a. solving the variability prognosity of the features, and selecting the features with changes for training the model, wherein the prognosity formula is as follows:
x j a measurement vector representing a feature on the jth system, the variable M being the number of monitored systems, N j Is the number of measurements on the j-th system, and for some features, if its prognosity is equal to 0 or NaN, these features are removed;
b. data normalization was performed using z-score:
u represents the mean value of the selected feature, sigma represents the standard deviation of the selected feature, and x represents a certain selected feature value;
c. dividing a time window, the window width being denoted as N t The sliding stride is denoted s, the first time window input1= [ x ] 1 ,x 2 ,...,x Nf ],x i A certain time feature vector is represented and,representing a temporal feature vector x i At the first of the time windowsValues, and so on;
a second time window, i.e., instant characteristic difference value Input 2 =[d 1 ,d 2 ,...,d Nf ],d i Representing the difference of a certain time feature vector;
B. dual channel long and short term memory network processing Input 1 And Input2, two outputs Output 1= [ h1, h2, ], h hidden_size ]And Output 2 =[g 1 ,g 2 ,...,g hidden_size ],h i Representing long and short term memory network processing Input 1 Vector g i Representing long and short term memory network processing Input 2 The vector of the following;
will Output 1 After (N) t -1) line vectors and Output 2 Direct addition, output is obtained, output= [ o ] 1 ,o 2 ,...,o hidden_size ],o i From h i And g i Adding to obtain;
C. the service life prediction is divided into two parts, wherein one part is to extract local time characteristics of a double-channel long-short-time memory network sequence output result by using a convolutional neural network, and the other part is to predict the residual service life by using a fully-connected neural network;
D. considering the buffer relation of the residual service life of the turbofan engine in the earlier stage to the residual service life of the current moment, inspiring the reduction of the momentum gradient, adopting a momentum smoothing residual service life method for a test set, and adopting the following formula:
predict t =k×y t +(1-k)×predict t-1 ,0≤k≤1(8)
y t the residual service life at the time t is predicted by using a double-channel long-short-time memory network t-1 Is the prediction result after the previous moment is smoothed, and predicts t Is the remaining service life after the current t moment is smoothed, k represents y t In the prediction t The proportion of the components;
2. the turbofan engine remaining life prediction method based on the dual-channel long-short-term memory network of claim 1, wherein the adopted data set can be divided into four sub-data sets FD001, FD002, FD003 and FD004, and each data set comprises a training set and a test set, and the details are shown in table 1:
table 1 data set introduction
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111053571.8A CN113722833B (en) | 2021-09-09 | 2021-09-09 | Turbofan engine residual service life prediction method based on double-channel long-short-term memory network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111053571.8A CN113722833B (en) | 2021-09-09 | 2021-09-09 | Turbofan engine residual service life prediction method based on double-channel long-short-term memory network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113722833A CN113722833A (en) | 2021-11-30 |
CN113722833B true CN113722833B (en) | 2023-06-06 |
Family
ID=78682835
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111053571.8A Active CN113722833B (en) | 2021-09-09 | 2021-09-09 | Turbofan engine residual service life prediction method based on double-channel long-short-term memory network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113722833B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114547803B (en) * | 2022-02-28 | 2023-04-18 | 扬州中卓泵业有限公司 | Ceramic pump turbine life detection system and method |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113094822A (en) * | 2021-03-12 | 2021-07-09 | 华中科技大学 | Method and system for predicting residual life of mechanical equipment |
CN113158445A (en) * | 2021-04-06 | 2021-07-23 | 中国人民解放军战略支援部队航天工程大学 | Prediction algorithm for residual service life of aero-engine with convolution memory residual self-attention mechanism |
Family Cites Families (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB0614114D0 (en) * | 2006-07-15 | 2006-08-23 | Rolls Royce Plc | An actuator |
CN102855349B (en) * | 2012-08-06 | 2015-07-01 | 南京航空航天大学 | Quick prototype design method and platform for gas path fault diagnosis for aeroengine |
US9605559B2 (en) * | 2015-02-02 | 2017-03-28 | General Electric Company | Wash timing based on turbine operating parameters |
CN106065830B (en) * | 2016-06-01 | 2017-11-24 | 南京航空航天大学 | A kind of pulse detonation combustor device combined based on rotary valve with pneumatic operated valve |
CN106524223B (en) * | 2016-12-15 | 2023-06-02 | 内蒙古中科朴石燃气轮机有限公司 | Combustion chamber with main nozzle assembly and mini-nozzle assembly |
CA3072045A1 (en) * | 2017-08-02 | 2019-02-07 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for detection in an industrial internet of things data collection environment with large data sets |
CN112703457A (en) * | 2018-05-07 | 2021-04-23 | 强力物联网投资组合2016有限公司 | Method and system for data collection, learning and machine signal streaming for analysis and maintenance using industrial internet of things |
CN112295145A (en) * | 2019-07-24 | 2021-02-02 | 上海长智系统集成有限公司 | Artificial intelligent fire-fighting robot, disaster detection method, computer device and medium |
CN112613226B (en) * | 2020-12-10 | 2022-11-18 | 大连理工大学 | Feature enhancement method for residual life prediction |
CN112966355B (en) * | 2021-03-30 | 2023-01-06 | 西安电子科技大学 | Method for predicting residual service life of shield machine cutter based on deep learning |
CN113204921B (en) * | 2021-05-13 | 2022-04-08 | 哈尔滨工业大学 | Method and system for predicting remaining service life of airplane turbofan engine |
-
2021
- 2021-09-09 CN CN202111053571.8A patent/CN113722833B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113094822A (en) * | 2021-03-12 | 2021-07-09 | 华中科技大学 | Method and system for predicting residual life of mechanical equipment |
CN113158445A (en) * | 2021-04-06 | 2021-07-23 | 中国人民解放军战略支援部队航天工程大学 | Prediction algorithm for residual service life of aero-engine with convolution memory residual self-attention mechanism |
Also Published As
Publication number | Publication date |
---|---|
CN113722833A (en) | 2021-11-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112149316B (en) | Aero-engine residual life prediction method based on improved CNN model | |
CN109472110B (en) | Method for predicting residual service life of aeroengine based on LSTM network and ARIMA model | |
CN110555479B (en) | Fault feature learning and classifying method based on 1DCNN and GRU fusion | |
CN110222371B (en) | Bayes and neural network-based engine residual life online prediction method | |
CN113722985B (en) | Method and system for evaluating health state and predicting residual life of aero-engine | |
CN110222901A (en) | A kind of electric load prediction technique of the Bi-LSTM based on deep learning | |
Liu et al. | Complex engineered system health indexes extraction using low frequency raw time-series data based on deep learning methods | |
CN113722833B (en) | Turbofan engine residual service life prediction method based on double-channel long-short-term memory network | |
CN113869563A (en) | Method for predicting remaining life of aviation turbofan engine based on fault feature migration | |
CN114266201B (en) | Self-attention elevator trapping prediction method based on deep learning | |
CN115017826A (en) | Method for predicting residual service life of equipment | |
Li et al. | A sequence-to-sequence remaining useful life prediction method combining unsupervised LSTM encoding-decoding and temporal convolutional network | |
CN111126477A (en) | Learning and reasoning method of hybrid Bayesian network | |
Xu et al. | Global attention mechanism based deep learning for remaining useful life prediction of aero-engine | |
CN113673774A (en) | Aero-engine remaining life prediction method based on self-encoder and time sequence convolution network | |
Li et al. | Remaining useful life prediction of aero-engine based on PCA-LSTM | |
Yao et al. | RUL prediction method for rolling bearing using convolutional denoising autoencoder and bidirectional LSTM | |
CN115048873B (en) | Residual service life prediction system for aircraft engine | |
CN116521406A (en) | Method for detecting anomaly of non-overrun flight parameter data of aero-engine based on residual gate GRU-VAE model | |
Wenqiang et al. | Remaining useful life prediction for mechanical equipment based on temporal convolutional network | |
CN115456287A (en) | Long-and-short-term memory network-based multi-element load prediction method for comprehensive energy system | |
CN113971489A (en) | Method and system for predicting remaining service life based on hybrid neural network | |
Yan et al. | Remaining Useful Life Interval Prediction for Complex System Based on BiGRU Optimized by Log-Norm | |
CN111008661A (en) | Croston-XGboost prediction method for reserve demand of aircraft engine | |
CN114741948B (en) | Aero-engine degradation trend prediction method based on residual stacking convolution network of sequence reconstruction |
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 |