CN114547986A - Data enhancement-based prediction method for residual life of aircraft engine - Google Patents

Data enhancement-based prediction method for residual life of aircraft engine Download PDF

Info

Publication number
CN114547986A
CN114547986A CN202210188342.5A CN202210188342A CN114547986A CN 114547986 A CN114547986 A CN 114547986A CN 202210188342 A CN202210188342 A CN 202210188342A CN 114547986 A CN114547986 A CN 114547986A
Authority
CN
China
Prior art keywords
data
prediction
data enhancement
model
features
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
Application number
CN202210188342.5A
Other languages
Chinese (zh)
Inventor
赵德群
赵嘉宇
张雅栋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Technology
Original Assignee
Beijing University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing University of Technology filed Critical Beijing University of Technology
Priority to CN202210188342.5A priority Critical patent/CN114547986A/en
Publication of CN114547986A publication Critical patent/CN114547986A/en
Pending legal-status Critical Current

Links

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
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
    • 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 invention discloses a method for predicting the residual life of an aircraft engine based on data enhancement, and provides a multipath feature fusion network model and a data enhancement algorithm. Firstly, the sample number of a data set is enlarged through a data enhancement mode, so that the accuracy of prediction is improved; secondly, constructing a multi-path feature fusion prediction model, and selecting two different paths to extract features: the first path inputs data into a Convolutional Neural Network (CNN) and a gating cycle unit (GRU) and respectively extracts spatial features and time sequence features; the second path inputs data directly into a long short term memory network (LSTM) to obtain timing characteristics. And finally, fusing the output characteristics of the two paths and inputting the fused output characteristics into a full connection layer to carry out RUL prediction. Compared with the existing network model, the method can effectively improve the residual life prediction precision of the equipment, and has practical application value.

Description

Data enhancement-based prediction method for residual life of aircraft engine
Technical Field
The invention relates to a method for predicting the residual life of an aircraft engine, in particular to a method for predicting the residual life of the aircraft engine based on a data enhancement and multi-path feature fusion model.
Background
Turbofan engines, as the "heart" of the various aircraft systems in the field of aviation, can directly affect the safe operation of the aircraft as a result of performance changes. However, as the aviation turbofan engine works in high-temperature, high-pressure and high-vibration-speed environments daily, accidents are easy to happen and faults are easy to occur along with the increase of working time. Therefore, the method has great significance for predicting the residual service life (RUL) of the aviation turbofan engine, changing the regular maintenance into the active maintenance and reducing the flight safety accidents and saving the equipment maintenance cost.
Currently, the remaining life prediction methods can be roughly divided into two categories, namely, prediction based on a physical model and prediction based on data driving. Due to the fact that the mechanical system is higher and higher in complexity and the parts are highly coupled, an accurate degradation model is difficult to establish by a prediction method based on a physical model, and flexibility and transportability are poor. The prediction method based on data driving analyzes collected equipment degradation data and fault data by utilizing technologies such as signal processing and the like, and digs out characteristics reflecting system degradation and faults so as to carry out RUL prediction. Among data-driven methods, machine learning and deep learning are widely used. The machine learning model is simple in structure, poor in deep level feature extraction capability on complex nonlinear multidimensional samples and low in prediction accuracy. In recent years, deep learning methods are gradually developed in the prediction field, and a Convolutional Neural Network (CNN) has strong learning capability and can well extract spatial features of data, but convolution and pooling operations can cause data loss. A Recurrent Neural Network (RNN) is a network having a unique advantage in processing time-series data, but RNN has a problem of gradient explosion or gradient disappearance in long-term prediction. Long short term memory networks (LSTM) alleviate the RNN extreme gradient problem by means of gate structures. And a gate control cycle unit (GRU) reduces a gate structure on the basis of the LSTM and improves the calculation efficiency. Most of the existing deep learning models are built based on a single model, the characteristics in multiple aspects cannot be mined and comprehensively considered, and the generalization performance is poor. In addition, in the aspect of degraded data, the existing performance degraded data set is less, and the requirement of a deep learning model on a large amount of training data cannot be met, so that the existing deep learning life prediction model is low in precision.
Disclosure of Invention
Aiming at the defects in the background technology, the invention provides a data-enhancement-based prediction method for the residual life of an aircraft engine.
The invention comprises the following steps:
step 1: and performing data enhancement and normalization on the acquired data by using a data enhancement algorithm.
Step 2: constructing a multi-path feature fusion prediction model, inputting the data processed in the step (1) into the model, and extracting different features in the data;
and step 3: fusing the features extracted in the step (2), and inputting the features into a full-connection layer to predict the final residual life;
and 4, step 4: and evaluating the model by adopting two evaluation methods, verifying the effectiveness of data enhancement, and comparing with other models.
Further, in step 1:
step 1.1: the acquired data is subjected to data enhancement processing by using a data enhancement algorithm, so that the requirement of deep learning on large data volume is met. The data enhancement algorithm is to generate partial training tracks by using a complete training track of a certain engine, wherein each partial training track generates a random point in the linear degradation process of the complete training track, the complete training track is cut off at the point, and the data enhancement is realized after the obtained partial training tracks are spliced to the complete training track. The partial training track generated by the method represents the performance track of the equipment before reaching the fault point, the track is closer to the test data, the quantity of the training data can be increased, and the test data can be better simulated.
Step 1.2: after step 1.1, the enhanced data is subjected to non-dimensionalization processing by using a z-score method, the sizes of various parameters in the data are limited to the same interval, and the z-score method is defined as follows:
Figure BDA0003524510280000021
in the formula, muiAnd σiRespectively represent the ith transmissionMean and standard deviation of sensor data.
Further, in step 2:
step 2.1: and selecting three network structures of CNN, GRU and LSTM to construct a multi-path feature fusion prediction model, and performing feature extraction on the performance degradation data by using different network advantages of the multi-path feature fusion prediction model.
Step 2.2: the CNN and the GRU form a first path of a model, and the CNN is used for extracting spatial features of the degraded data, and then the spatial features are input into the GRU to extract time sequence features. Since the CNN performs convolution and pooling operations during feature extraction, which may result in data loss, the LSTM is used as a second path of the model to separately extract the time-series features of the performance-degraded data. Because the data is enhanced in step 1, the GRU with the fast calculation speed is selected to extract the time sequence feature of the first path. In the model, a Dropout mechanism is used for reducing overfitting and improving the fitting effect of the model.
Further, in step 3:
step 3.1: and (3) fusing the features respectively extracted from the two paths in the multi-path feature fusion prediction model in the step (2) by adopting a concat method in Tensorflow to obtain the space-time features.
Step 3.2: and (3) inputting the fused space-time characteristics in the step (3.1) into two full-connection layers for operation, wherein the last full-connection layer only has one output value, namely the predicted value of the remaining life of the engine.
Further, in step 4:
the predictive power of the model was evaluated by evaluating the predicted remaining life in step 3 using Root Mean Square Error (RMSE) and scoring function (Score). RMSE evaluates the ability of the model to estimate unbiased, and Score increases the penalty weight of the lag prediction, and the expressions of RMSE and Score are as follows:
Figure BDA0003524510280000031
in the formula, EiError representing the ith prediction, Ei<0 represents a superPre-prediction, Ei>0 denotes a lag prediction.
And finally, evaluating the data subjected to data enhancement and the data not subjected to data enhancement by utilizing two evaluation methods of RMSE and Sorce, verifying the effectiveness of data enhancement, and comparing the multi-path characteristic fusion prediction model with other prediction models in the aspects of RMSE and Sorce.
Compared with the prior art, the invention has the following beneficial effects:
the number of model training data is increased through a data enhancement algorithm, and the requirement of deep learning on large data volume is met; part of training track data generated in the data enhancement algorithm is similar to test data in form, the prediction performance of the model can be enhanced in the model training process, and the prediction accuracy is improved.
The multi-path feature fusion prediction model integrates the advantages of various algorithms into one model through a parallel structure, and simultaneously learns local spatial features and time sequence features to realize the full mining of input data by a network, thereby achieving the aim of high-precision prediction; by combining different advantages of the CNN and the RNN, the number of parameters is reduced, the calculation speed is further increased, the calculation complexity is reduced, and the overfitting phenomenon in training is reduced.
Drawings
FIG. 1 is a data enhancement; (a) no data enhancement; (b) and (4) enhancing data.
FIG. 2 is a multi-path feature fusion prediction model structure.
FIG. 3 is a flow chart of a multi-path feature fusion prediction model.
FIG. 4 shows the RUL prediction results.
Detailed Description
The method for predicting the residual life of the aircraft engine based on data enhancement is further described below with reference to the accompanying drawings.
The invention discloses a method for predicting the residual life of an aircraft engine based on a data enhancement method and a multi-path characteristic fusion prediction model by adopting a turbofan engine degradation simulation data set disclosed by NASA.
The method comprises the following steps: data prediction processing
And carrying out data enhancement and data normalization processing on the adopted CMAPSS data set.
Data enhancement detailed description as shown in fig. 1, (a) of fig. 1 is a complete training trajectory of an engine RUL, where each time in the trajectory contains multidimensional data. In the enhancement algorithm, a partial training trajectory is generated using the complete training trajectory. Three partial training trajectories are generated as shown in fig. 1 (b) using the full training trajectory in fig. 1 (a). Each part of training track generates a random point in the linear degradation process, the complete training track is cut off at the random point, and the obtained part of training track is placed behind the complete training track to achieve the purpose of data enhancement. The generated partial training trajectory represents the performance trajectory of the device before reaching the failure point, and the trajectory is closer to the data in the test set. The data enhancement method can increase the data volume of the training set and better simulate the data of the test set, so that the model can better learn the data characteristics and improve the prediction accuracy of the model.
After data enhancement, carrying out non-dimensionalization processing on the enhanced data by adopting a z-score method, limiting the sizes of various parameters in the data in the same interval, and preventing the influence on a prediction result, wherein the z-score method is defined as follows:
Figure BDA0003524510280000041
in the formula, muiAnd σiRespectively, mean and standard deviation of the ith sensor data.
Step two: and constructing a multi-path feature fusion prediction model, and extracting hidden features in the data.
CNN, GRU and LSTM are typical models for solving the problem of time series at present, and the multi-path feature fusion prediction model has the unique advantages of various models, and is structurally shown in figure 2 and consists of two parallel paths and a full connection layer. The first path consists of CNN and GRU, and the data is pre-processed and input into CNN to extract spatial features and then input into GRU to extract time sequence features. Since CNN performs feature extraction using convolution and pooling operations, part of information is lost in this process, resulting in incomplete raw data timing feature extraction by GRU, and therefore a second parallel path composed of LSTM is added to extract complete timing features. And finally, fusing the extracted data characteristics into space-time characteristics to carry out final RUL prediction. For ease of understanding, the proposed flow chart is shown in fig. 3.
Step three: feature fusion
And (5) fusing the features respectively extracted from the two paths in the multi-path feature fusion prediction model in the step two by adopting a concat () method in Tensorflow to obtain space-time features. And inputting the fused space-time characteristics into two full-connection layers for operation, wherein the last full-connection layer only has one neuron, and the output result is a predicted value of the remaining life of the engine.
Step four: and evaluating the model, verifying the effectiveness of data enhancement, and comparing with other models.
The present invention uses two evaluation criteria, Root Mean Square Error (RMSE) and scoring function (Score). Nth prediction error EnRMSE and Score expressions are as follows:
En＝RULEst-RULTrue
RULEstand RULTrueRespectively representing predicted and true values, Ei<0 denotes advance prediction, Ei>0 represents a lag predictor that may create a safety hazard in practice, and therefore, a scoring function is employed that imposes a large penalty on the lag prediction:
Figure BDA0003524510280000051
the scoring function is sensitive to abnormal values, and one abnormal value can greatly change the value of the scoring function because the prediction error is not normalized, so that the RMSE is adopted to evaluate the unbiased estimation capability of the algorithm:
Figure BDA0003524510280000052
the data subjected to data enhancement and the data not subjected to data enhancement are input into a multi-path feature fusion prediction model after being normalized respectively, the residual life of the engine is predicted, two evaluation indexes, namely RMSE and Score, are used for verifying the performance, the result is shown in table 1, and the result shows that the evaluation index subjected to data enhancement and prediction is obviously superior to the prediction result not subjected to data enhancement in numerical value, so that the prediction performance is improved. The data enhancement method has positive influence on life prediction, so that the model fitting degree is better to a certain degree, and the prediction result is more accurate.
Table 1 data enhancement results comparison
Figure BDA0003524510280000053
The invention uses the test data in the data set to verify the model performance, and fig. 4 shows the RUL prediction result of the engine No. 68 in the test set, and as can be seen from the figure, the engine No. 68 shows a relatively stable state in the whole test, and the predicted value curve is closely attached to the actual value curve. Although the test may fluctuate earlier, the prediction curve gradually tends to be smooth as the engine operating cycle continues to increase.
The multi-path feature fusion prediction model of the present invention was compared to other RUL prediction models. Because the model prediction has certain randomness, the method carries out a plurality of experiments to obtain the average result of RMSE and Source. As shown in table 2, the method used herein achieves better overall results than other methods.
As can be seen from Table 2, the model structure of the present invention achieves the lowest values in both RMSE and Score, indicating that the model structure herein has higher prediction accuracy and obtains better performance in turbofan engine prediction. Compared with a single LSTM, the RMSE values in the 4 subsets of the model were reduced by 26.19%, 25.19%, 19.78%, 32.02%, respectively. Compared with CNN-GRU, the RMSE of the model of the invention is reduced by 35.4%, 32.54%, 30.35% and 29.36%, respectively. The model of the invention also obtains a lower Score value in prediction, and shows the advance of the prediction result, which has higher safety and effectiveness in the application of actual prediction and health management.
TABLE 2 comparison of different prediction methods
Figure BDA0003524510280000061
The method for predicting the residual life of the aircraft engine based on data enhancement provided by the invention is introduced, the principle and the implementation mode of the invention are explained, and the description of the embodiment is only used for helping to understand the method and the core idea of the invention; it will be apparent to those skilled in the art that changes in the embodiments and applications may be made without departing from the spirit of the invention, and the invention is not to be considered limited to the details set forth in the specification.

Claims (10)

1. A method for predicting the residual life of an aircraft engine based on data enhancement is characterized by comprising the following steps:
step 1: enhancing the training data by using a data enhancement algorithm, and normalizing the training data and the test data;
step 2: constructing a multi-path feature fusion prediction model, inputting the data processed in the step (1) into the model, and extracting different features in the data;
and step 3: fusing the features extracted in the step (2), and inputting the features into a full-connection layer to predict the final residual life;
and 4, step 4: and evaluating the model by adopting two evaluation methods, verifying the effectiveness of data enhancement, and comparing with other models.
2. The method for predicting the remaining life of the aero-engine based on data enhancement as claimed in claim 1, wherein in step 1, the data enhancement algorithm utilizes a complete training trajectory of an engine to perform data enhancement.
3. The method for predicting the remaining life of the aero-engine based on data enhancement as claimed in claim 2, wherein partial training trajectories are generated by using a complete training trajectory of the engine, each partial training trajectory is a random point generated in a linear degradation process of the complete training trajectory, the complete training trajectory is cut off at the point, and data enhancement is achieved after the obtained partial training trajectories are spliced to the complete training trajectory.
4. The method for predicting the remaining life of the aero-engine based on data enhancement as claimed in claim 3, wherein the generated part of the training track represents a performance track of the equipment before reaching the fault point, the performance track is closer to the test data, the number of the training data can be increased, and the test data can be better simulated.
5. The method for predicting the residual life of the aircraft engine based on the data enhancement is characterized in that in the step 1, a z-score method is adopted for normalization, and the expression is as follows:
Figure FDA0003524510270000011
in the formula, muiAnd σiRespectively, mean and standard deviation of the ith sensor data.
6. The method for predicting the residual life of the aircraft engine based on data enhancement as claimed in claim 1, wherein in step 2, a multipath feature fusion prediction model is constructed by using CNN, GRU and LSTM, and feature extraction is carried out on input data.
7. The method as claimed in claim 6, wherein CNN and GRU are the first path of the model, CNN extracts spatial features from the input data and then inputs GRU to extract time series features of the data, and LSTM is the second path of the model and extracts time series features of the input data.
8. The method for predicting the residual life of the aero-engine based on data enhancement as claimed in claim 1, wherein in step 3, features on two paths extracted from the multi-path feature fusion prediction model in step 2 are fused by a concat () method in Tensorflow, and the fused features are input into two full-connection layers for residual life prediction after being fused into space-time features.
9. The method of claim 1, wherein in step 4, the predicted remaining life value of step 3 is evaluated using Root Mean Square Error (RMSE) and a scoring function (Score), the RMSE evaluating the ability of the model to estimate unbiased, the Score weighting the penalty of lag prediction is increased, and the RMSE and Score are expressed as follows:
Figure FDA0003524510270000021
in the formula, EiError representing the ith prediction, Ei<0 denotes advance prediction, Ei>0 denotes a lag prediction.
10. The method for predicting the residual life of the aircraft engine based on the data enhancement is characterized in that in the step 4, data which are subjected to data enhancement and data enhancement which are not subjected to data enhancement are evaluated by utilizing two evaluation methods of RMSE and Sorce, the effectiveness of the data enhancement is verified, and the multi-path feature fusion prediction model is compared with other prediction models in terms of RMSE and Sorce.
CN202210188342.5A 2022-02-28 2022-02-28 Data enhancement-based prediction method for residual life of aircraft engine Pending CN114547986A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210188342.5A CN114547986A (en) 2022-02-28 2022-02-28 Data enhancement-based prediction method for residual life of aircraft engine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210188342.5A CN114547986A (en) 2022-02-28 2022-02-28 Data enhancement-based prediction method for residual life of aircraft engine

Publications (1)

Publication Number Publication Date
CN114547986A true CN114547986A (en) 2022-05-27

Family

ID=81679833

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210188342.5A Pending CN114547986A (en) 2022-02-28 2022-02-28 Data enhancement-based prediction method for residual life of aircraft engine

Country Status (1)

Country Link
CN (1) CN114547986A (en)

Similar Documents

Publication Publication Date Title
CN112131760B (en) CBAM model-based prediction method for residual life of aircraft engine
Li et al. A directed acyclic graph network combined with CNN and LSTM for remaining useful life prediction
CN112580263B (en) Turbofan engine residual service life prediction method based on space-time feature fusion
CN113158445B (en) Prediction algorithm for residual service life of aero-engine with convolution memory residual error self-attention mechanism
CN110609524B (en) Industrial equipment residual life prediction model and construction method and application thereof
CN114266278B (en) Dual-attention network-based equipment residual service life prediction method
CN115017826B (en) Method for predicting residual service life of equipment
CN113869563A (en) Method for predicting remaining life of aviation turbofan engine based on fault feature migration
CN114880925A (en) Equipment life prediction method based on time convolution network and multi-layer self-attention
CN112881987A (en) Airborne phased array radar behavior prediction method based on LSTM model
CN116662743A (en) Engine residual life prediction method based on multi-mode deep learning
CN114169091A (en) Method for establishing prediction model of residual life of engineering mechanical part and prediction method
CN115048873B (en) Residual service life prediction system for aircraft engine
CN114547986A (en) Data enhancement-based prediction method for residual life of aircraft engine
CN112560252B (en) Method for predicting residual life of aeroengine
CN114841063A (en) Aero-engine residual life prediction method based on deep learning
CN112100767A (en) Aero-engine service life prediction method based on singular value decomposition and GRU
Cui et al. Prediction of Aeroengine Remaining Useful Life Based on SE-BiLSTM
Li et al. Application of Improved PSO-BP neural network in fault detection of liquid-propellant rocket engine
Zhou et al. Remaining Useful Life Prediction of Aero-Engine using CNN-LSTM and mRMR Feature Selection
Lu et al. Remaining Useful Life Prediction and Health Status Estimation Based on Joint-Loss Convolution Neural Networks
Yang et al. A Novel Transfer Learning Method Based on Domain Adversarial Networks for Remaining Useful Life Prediction
Yang et al. Remaining Useful Life Prediction Based on Stacked Sparse Autoencoder and Echo State Network
Qin et al. Aeroengine Life Prediction Based on CNN-BiLSTM Model and Attention Mechanism
Feng et al. Remaining useful life estimation of aeroengine based on MPFAM-FCN

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