CN113836822A - Aero-engine service life prediction method based on MCLSTM model - Google Patents

Aero-engine service life prediction method based on MCLSTM model Download PDF

Info

Publication number
CN113836822A
CN113836822A CN202111261118.6A CN202111261118A CN113836822A CN 113836822 A CN113836822 A CN 113836822A CN 202111261118 A CN202111261118 A CN 202111261118A CN 113836822 A CN113836822 A CN 113836822A
Authority
CN
China
Prior art keywords
model
data
mclstm
inputting
output
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
CN202111261118.6A
Other languages
Chinese (zh)
Inventor
秦毅
项盛
周弦
罗均
周江洪
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.)
Chongqing University
Original Assignee
Chongqing University
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 Chongqing University filed Critical Chongqing University
Priority to CN202111261118.6A priority Critical patent/CN113836822A/en
Publication of CN113836822A publication Critical patent/CN113836822A/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
    • 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
    • G06N3/047—Probabilistic or stochastic 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
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Business, Economics & Management (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides an aircraft engine service life prediction method based on an MCLSTM model, which comprises the following steps: acquiring multidimensional degradation parameters of an aircraft engine to be predicted to obtain acquired data, dividing the acquired data by adopting a sliding window technology to obtain preprocessed data, constructing a life prediction model comprising an MCLSTM model and a statistical model, inputting the preprocessed data into the MCLSTM model and the statistical model respectively to obtain health indexes MHI and THI based on the MCLSTM model, connecting the MHI and the THI to form a health index data set, and inputting the health index data set into a regression layer of the life prediction model to predict the residual service life. One branch of the life prediction model extracts MHI from input data through the MCLSTM model, the other branch extracts THI from the input data through the statistical model, and finally indexes built based on the two branches are based on regression prediction.

Description

Aero-engine service life prediction method based on MCLSTM model
Technical Field
The invention relates to the field of prediction of the service life of an aero-engine, in particular to a method for predicting the service life of an aero-engine based on an MCLSTM model.
Background
The aircraft engine is a highly complex and precise thermal machine, is an engine for providing power required by flight for an aircraft, and is more prone to failure due to a complex internal structure and a severe working environment; therefore, the accurate prediction of the residual service life of the aircraft engine is greatly helpful to the operation and maintenance of the aircraft engine.
The existing neural network prediction models have the problems that different updating modes can not be carried out on different input data according to the importance degree of the input data, the model calculation amount is large, and the prediction accuracy of the model is poor.
Disclosure of Invention
The invention aims to provide an aircraft engine life prediction method based on an MCLSTM model, which can be used for predicting the residual life of an aircraft engine.
The invention is realized by the technical scheme, and the method comprises the following specific steps:
1) data acquisition: acquiring multidimensional degradation parameters of the aero-engine to be predicted, and selecting a plurality of parameters capable of reflecting the degradation performance of the aero-engine through stability trend analysis to obtain acquired data;
2) data preprocessing: the collected data is divided by adopting a sliding window technology to obtain preprocessed data xi；
3) Constructing a model: constructing an aircraft engine life prediction model comprising a multicellular long-term and short-term memory neural network MCLSTM model and a statistical model;
4) extracting health indexes: preprocessing data xiRespectively inputting the health indexes into an MCLSTM model and a statistical model to obtain a health index MHI based on the MCLSTM model and a health index THI based on the statistical model;
5) and (3) life prediction: and connecting the MHI and the THI to form a health index data set, and inputting the health index data set into a regression layer of the life prediction model of the aircraft engine to predict the residual service life.
Further, the prediction model of the service life of the aircraft engine comprises an MCLSTM model, a statistical model, a full connection layer FC1, a full connection layer FC2, a full connection layer FC3 and a regression layer, wherein the MCLSTM model comprises a hierarchical division unit and a multi-cell updating unit.
Further, the specific steps of extracting the health index MHI based on the MCLSTM model in the step 4) are as follows:
4-1-1) inputting the preprocessed data in the step 2) into an MCLSTM model as input data, performing hierarchical division on the input data through a hierarchical division unit, and updating the data after hierarchical division through a multi-cell updating unit to obtain hidden features;
4-1-2) inputting the obtained hidden features into an attention mechanism to obtain attention weights, and combining the attention weights with the hidden features; obtaining a merging characteristic;
4-1-3) inputting the merged characteristics into the full connection layer FC1 to obtain the output of the full connection layer FC1, and inputting the output of the full connection layer FC1 into the full connection layer FC2 to obtain the health index MHI based on the MCLSTM model.
Further, the method for extracting the health index THI based on the statistical model in the step 4) comprises the following steps:
inputting the preprocessed data in the step 2) into a statistical model, counting the mean value and the trend coefficient of the input data through the statistical model, inputting the statistical data into a full-connection layer FC3 of an aircraft engine service life prediction model, and obtaining a health index THI based on the statistical model.
Further, the specific steps of obtaining the hidden features in the step 4-1-1) are as follows:
4-1-1-1) hierarchical partitioning: taking the preprocessed data in the step 2) as input data, and taking the input data x at the time tt＝[xt,1 xt,2 … xt,n]TAnd recursive data h of the MCLSTM model at time t-1t-1＝[ht-1,1 ht-1,2 … ht-1,m]TInputting the output A into a full connection layer of a multi-cell long-short term memory neural network MCLSTM model, activating the output through a tanh activation function to obtain the output A of the full connection layer, and regularizing the output A of the full connection layer through a SoftMax function to obtain the output A of a hierarchical division unit1：
A＝tanh(wxaxt+whaht-1+ba) (1)
A1＝softmax(A)＝[a1 a2 a3 a4 a5] (2)
In the formulae (1) and (2), wxaAnd whaRespectively input data weight and recursive data weight in the full link layer, baFor full link layer bias, a1,a2,a3,a4And a5Respectively representing five levels, high, medium, low and medium, A1The element position of the middle maximum element represents the level of data input into the MCLSTM at the time t, and n and m are the characteristic dimensions of the input data and the recursion data respectively;
4-1-1-2) extracting hidden features: inputting data incapable of hierarchy into the multi-cell update unit c according to the hierarchy division structuretFive different dissimilatory units ct(m)、ct(i)、ct(j)、ct(k)、ct(l) According to the update results of five different units and the hierarchy division result A1The output c of the multi-cell renewal unit is obtained by weighting and combining the differentiation outputs of the sub-cell units in the multi-cell renewal unittAnd extracting hidden feature h at time tt：
Figure BDA0003325770580000031
In the formula (3), ct(m) is high level data a1Corresponding data update unit, ct(l) For medium and high level data a2Corresponding data update unit, ct(k) Is composed ofHierarchical data a3Corresponding data update unit, ct(j) For medium and low level data a4Corresponding data update unit, ct(i) As low-level data a5Corresponding data update unit, sigma is sigmod activation function, wixAs a weight between the input data and the MCLSTM model input gate, wihFor the weight between the recursive data and the input gate of the MCLSTM model, wfxAs a weight between the input data and the MCLSTM model forget gate, wfhIs the weight between the recursive data and the MCLSTM model forget gate, woxAs a weight between the input data and the output gate of the MCLSTM model, wohAs weights between the recursive data and the output gates of the MCLSTM model, biInput gate bias for MCLSTM model, bfForget gate bias for MCLSTM model, boOutput gate offset for MCLSTM model, ct-1Indicates the memory cell output, w, at time t-1cxAs a weight between the input data and the cell memory unit, wchWeights between recursive data and cell memory units, bcBias for cell memory cell in MCLSTM model, <' > is dot product operation, s1、s2All are mixing scale coefficients obtained by learning.
Due to the adoption of the technical scheme, the invention has the following advantages:
1. according to the intelligent life prediction model for the aircraft engine, one branch of the intelligent life prediction model extracts the health index MHI based on the MCLSTM model from input data and working condition information through the MCLSTM model, the other branch carries out statistics on the mean value and the trend coefficient of the input data through the statistical model to obtain the health index THI based on the statistical model, and finally regression prediction is carried out on the health index constructed based on the two branches.
2. The MCLSTM model determines the importance degree of input data through the hierarchical division unit, then the multi-cell updating unit with different data updating modes is designed to conduct differential updating on the input data, and the health state degradation trend of the aero-engine in different degrees can be better mined.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof.
Drawings
The drawings of the present invention are described below.
FIG. 1 is a flow chart of a method for predicting the life of an aircraft engine according to the present invention.
FIG. 2 is a schematic structural diagram of the MCLSTM model of the multicellular long-short term memory neural network of the present invention.
FIG. 3 is a diagram illustrating a sliding window technique in data preprocessing according to the present invention.
Fig. 4 is a diagram of the prediction result of the FD001 subset according to the prediction method of the present invention.
Fig. 5 is a diagram of the prediction result of the prediction method of the present invention on the FD004 subset.
Detailed Description
The invention is further illustrated by the following figures and examples.
The method for predicting the life of the aircraft engine based on the MCLSTM model, as shown in FIGS. 1-3, comprises the following specific steps:
1) data acquisition: collecting multidimensional degradation parameters of an aero-engine to be predicted, selecting a plurality of parameters capable of reflecting the degradation performance of the aero-engine through stable trend analysis, and obtaining collected data, wherein the method comprises the following specific steps:
1-1) simulating degradation data of an aircraft engine in a C-MAPSS (commercial modular aviation propulsion system simulation), and collecting multidimensional degradation parameters of the aircraft engine to be predicted, wherein the parameters are shown in a table 1:
TABLE 1 output of 21 Sensors when Engine is running
Figure BDA0003325770580000041
Figure BDA0003325770580000051
As shown in table 2, the C-MAPSS dataset was divided into 4 sub-datasets according to different operating conditions and failure modes:
TABLE 2C-MAPSS data set
Figure BDA0003325770580000052
Each subdata set comprises training data, test data and actual residual life RUL corresponding to the test data; the training data comprises all engine data from a certain health state to a fault, and the test data is data before the engine operation fault; the training data and the test data each contain a number of engines having different initial health states.
Due to the different initial health states of the engines, the operation periods of different engines in the same database are different. Taking the FD001 data set as an example, the training data set includes 100 engines, the maximum operation period is 362, and the minimum operation period is 128. In order to fully prove the superiority of the method, the simplest subset, namely the subset FD001 with single working condition and single fault, and the most complex subset, namely the subset FD004 with multiple working conditions and multiple fault modes are adopted as experimental data.
1-2) some stable trend measurements (measurement data of sensors 1, 5, 6, 10, 16, 18 and 19) were excluded in advance. These sensor measurements are not suitable for residual life RUL prediction because the full life measurement profile of these measurements is stable, constant, i.e. contains less engine degradation information, and because operating condition information has a significant impact on the predictive capabilities of the model. Therefore, the 14 screened sensor measurements and the working condition information jointly form raw data to obtain collected data.
2) Data preprocessing: the method comprises the following steps of adopting a sliding window technology to carry out data segmentation on collected data to obtain preprocessed data, and specifically comprises the following steps:
as shown in fig. 3, if the engine's full life cycle is T, the sliding window size is set to l, the sliding step size is set to m, and the size of the ith input sample is l × n, where: n is the sum of the number of the selected sensors and the working condition information dimension;
the actual residual life RUL is T-l- (i-1) x m when the ith sample is input;
the linear piecewise residual lifetime RUL technique is used to construct a lifetime label, which is defined as follows:
Figure BDA0003325770580000061
in formula (4), RulmaxA threshold value set in advance for the maximum remaining life;
the maximum life values of FD001 and FD004 are set to 130 and 150 operating cycles, respectively, the sliding window size l is set to 30, and the sliding step m is set to 1. The training sample sizes for FD001 and FD004 can be calculated as 17731 and 54028, respectively, and the number of test samples is 100248, respectively, because only the last measurement of the test set was used for prediction capability verification.
3) Constructing a model: the method comprises the steps of constructing an aero-engine life prediction model, wherein the aero-engine life prediction model comprises a multi-cell long-short term memory neural network (MCLSTM) model, a statistical model, a full connection layer FC1, a full connection layer FC2, a full connection layer FC3 and a regression layer, and the MCLSTM model comprises a hierarchical division unit and a multi-cell updating unit.
4) Extracting health indexes: preprocessing data xtRespectively inputting the health indexes into an MCLSTM model and a statistical model to obtain a health index MHI based on the MCLSTM model and a health index THI based on the statistical model, and the specific steps are as follows:
4-1) extracting the health index MHI based on the MCLSTM model:
4-1-1) inputting the preprocessed data in the step 2) into an MCLSTM model as input data, performing hierarchical division on the input data through a hierarchical division unit, and updating the hierarchically-divided data through a multi-cell updating unit to obtain a hidden feature htThe detailed stepsComprises the following steps:
4-1-1-1) hierarchical partitioning: taking the preprocessed data in the step 2) as input data, and taking the input data x at the time tt＝[xt,1 xt,2 … xt,n]TAnd recursive data h at time t-1t-1＝[ht-1,1 ht-1,2 ... ht-1,m]TInputting the output A into a full connection layer of a multi-cell long-short term memory neural network MCLSTM model, activating the output through a tanh activation function to obtain the output A of the full connection layer, and regularizing the output A of the full connection layer through a SoftMax function to obtain the output A of a hierarchical division unit1：
A＝tanh(wxaxt+whaht-1+ba) (5)
A1＝softmax(A)＝[a1 a2 a3 a4 a5] (6)
In the formulae (5) and (6), wxaAnd whaRespectively input data weight and recursive data weight in the full link layer, baFor full link layer bias, a1,a2,a3,a4And a5Respectively representing five levels, high, medium, low and medium, A1The element position of the middle maximum element represents the level of data input into the MCLSTM at the time t, and n and m are the characteristic dimensions of the input data and the recursion data respectively;
4-1-1-2) extracting hidden features: inputting data incapable of hierarchy into the multi-cell update unit c according to the hierarchy division structuretFive different dissimilatory units ct(m)、ct(i)、ct(j)、ct(k)、ct(l) According to the update results of five different units and the hierarchy division result A1The output c of the multi-cell renewal unit is obtained by weighting and combining the differentiation outputs of the sub-cell units in the multi-cell renewal unittAnd extracting hidden feature h at time tt：
Figure BDA0003325770580000071
In the formula (7), ct(m) is high level data a1Corresponding data update unit, ct(l) For medium and high level data a2Corresponding data update unit, ct(k) Is middle level data a3Corresponding data update unit, ct(j) For medium and low level data a4Corresponding data update unit, ct(i) As low-level data a5Corresponding data update unit, sigma is sigmod activation function, wixAs a weight between the input data and the MCLSTM model input gate, wihFor the weight between the recursive data and the input gate of the MCLSTM model, wfxAs a weight between the input data and the MCLSTM model forget gate, wfhIs the weight between the recursive data and the MCLSTM model forget gate, woxAs a weight between the input data and the output gate of the MCLSTM model, wohAs weights between the recursive data and the output gates of the MCLSTM model, biInput gate bias for MCLSTM model, bfForget gate bias for MCLSTM model, boOutput gate offset for MCLSTM model, ct-1Indicates the memory cell output, w, at time t-1cxAs a weight between the input data and the cell memory unit, wchWeights between recursive data and cell memory units, bcBias for cell memory cell in MCLSTM model, <' > is dot product operation, s1、s2All are mixing scale coefficients obtained by learning.
4-1-2) hidden feature h to be obtainedtInputting the attention weight into an attention mechanism, obtaining an attention weight, and combining the attention weight with the hidden feature; obtaining a merging characteristic;
4-1-3) inputting the merged characteristics into the full connection layer FC1 to obtain the output of the full connection layer FC1, and inputting the output of the full connection layer FC1 into the full connection layer FC2 to obtain the health index MHI based on the MCLSTM model.
In the present example, the high level data represents a global trend of aircraft engine degradation in the hope of obtaining a long-term retention to accurately assess the overall trend of aircraft engine degradationThus, subcellular unit ct(m) updating by preserving the memory cell state of the multicellular updating cells at the last moment.
The data at the high and high level represent the medium and long term trends in the process of aircraft engine degradation, which are input into subcellular unit ct(l) Updating and learning, namely proportionally mixing the medium-term degradation trend and the long-term degradation trend for updating ct(l)。
The mid-level data represents the metaphase trend in the degeneration process, and this data will be represented by subcellular unit ct(k) Update, the same as the cell unit update mode in the conventional long-short term memory neural network LSTM, ct(k) Consisting of a non-linear combination of long-term and short-term degradation trends.
The data at the medium and low level represent short-term trends in the aircraft engine degradation process, which are input into subcellular unit ct(j) And performing update learning. Likewise, the medium-term degradation tendency and the short-term degradation tendency are proportionally mixed for updating ct(j)。
The low-level data represents short-term fluctuation in the degradation process of the aircraft engine, and the low-level data is expected to be updated in real time so as to rapidly acquire local change of the degradation trend of the aircraft engine, namely the subcellular unit ct(i) The internal state c of this momenttAnd (6) updating.
4-2) extracting a health index THI based on a statistical model:
inputting the preprocessed data in the step 2) into a statistical model, counting the mean value and the trend coefficient of the input data through the statistical model, inputting the statistical data into a full-connection layer FC3 of an aircraft engine service life prediction model, and obtaining a health index THI based on the statistical model.
5) And (3) life prediction: and connecting the MHI and the THI to form a health index data set, and inputting the health index data set into a regression layer of the life prediction model of the aircraft engine to predict the residual service life.
In the embodiment of the present invention, the total number of samples in the training process is N, and the Mean Square Error (MSE) is defined as a loss function, and then the calculation formula is:
Figure BDA0003325770580000081
in the formula (8), the reaction mixture is,
Figure BDA0003325770580000082
and RuliRespectively predicting the residual life and the actual residual life of the ith sample; and obtaining the error gradient of each layer by using a back propagation rule, optimizing the weight parameter of the model by using an Adam optimization method, and avoiding the over-fitting problem in deep learning by using a Dropout technology.
Selecting hyper-parameters of the life prediction model of the aircraft engine as follows: the number of nerve cells of MCLSTM was 50, the number of nerve cells of full junction layer FC1 was 50, the number of nerve cells of full junction layers FC2 and FC3 was 10, the number of nerve cells of regression layer was 1, the learning rate was 0.001, and Dropout1 and Dropout2 were both set to 0.2.
6) And (3) experimental verification:
6-1) constructing an evaluation index: and (3) quantitatively characterizing the residual service life prediction performance by adopting an IEEE evaluation Score (Score) and a Root Mean Square Error (RMSE) as evaluation indexes, wherein the evaluation indexes can be respectively calculated as follows:
Figure BDA0003325770580000091
Figure BDA0003325770580000092
Figure BDA0003325770580000093
in formulae (9), (10), (11), RuliAnd
Figure BDA0003325770580000094
actual and predicted life of the ith engine, respectively, N being the total number of engines in the subset. The values of these indices are inversely proportional to the RUL performance, i.e. the lower the index value, the better the model performance. The Score crossing the RMSE penalizes over-prediction more than is the case in engineering practice. Therefore, with RMSE close, the model is more prone to be evaluated based on Score;
6-2) predicting and comparing the residual service life: in order to comprehensively evaluate the prediction capability of the proposed model and consider the problem of randomness distribution of initial weights in the deep learning model, 10 parallel experiments are performed by the proposed method based on FD001 and FD004 data, comparison is performed by several advanced methods, various latest methods such as a statistical method, a shallow machine learning model and a deep learning model are compared with the proposed deep learning method, and the advantages of the proposed method in the aspect of RUL prediction on two subsets are highlighted. The quantitative values of the evaluation indexes of different methods are shown in Table 3.
TABLE 3 quantitative comparison of predicted Performance on FD001 and FD004 by different methods
Figure BDA0003325770580000095
Overall, all methods performed better for residual life RUL prediction on FD001, since FD001 is simpler than FD 004. In addition, the number of engines of the FD004 is much larger than that of the FD001 due to the difference in the number of engines of the two subsets, and Score is an accumulated evaluation index. Thus, the scores for the same model in the two subsets are very different. As can be seen from Table 3, LSTM, DBN, MODENB and the attention-based deep learning methods all performed better than the shallow learning methods and statistical algorithms in terms of RUL prediction. Experimental results show that the deep learning method has a deeper structure, so that the deep learning method has a stronger feature extraction capability, and the performance of CNN is inferior to ELM and RF because CNN is not suitable for processing time series data. MODEBNE is used as an ensemble learning model based on DBN, and is superior to other DL methods by utilizing the advantages of feature extraction and ensemble learning of DBN. The attention-based deep learning approach performs best because of the use of manual features and attention mechanisms. The effectiveness of the deep learning framework used is illustrated, and the effectiveness of the MCLSTM is further verified. Due to the MCLSTM, the global trend can be obtained as much as possible and the local trend can be updated as soon as possible by adopting a deep learning method, and compared with other reference methods, the RUL prediction capability is improved.
As shown in fig. 4 and 5, the correspondence between the residual life RUL predicted by the method for predicting the life of an aircraft engine based on the MCLSTM model and the actual residual life RUL described in the present application indicates that the two residual life values approximately coincide with each other, so that the prediction of the residual life RUL of the proposed model is feasible.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.

Claims (5)

1. An aircraft engine life prediction method based on an MCLSTM model is characterized by comprising the following specific steps:
1) data acquisition: acquiring multidimensional degradation parameters of the aero-engine to be predicted, and selecting a plurality of parameters capable of reflecting the degradation performance of the aero-engine through stability trend analysis to obtain acquired data;
2) data preprocessing: performing data segmentation on the acquired data by adopting a sliding window technology to obtain preprocessed data;
3) constructing a model: constructing an aircraft engine life prediction model comprising a multicellular long-term and short-term memory neural network MCLSTM model and a statistical model;
4) extracting health indexes: respectively inputting the preprocessed data into the MCLSTM model and the statistical model to obtain a health index MHI based on the MCLSTM model and a health index THI based on the statistical model;
5) and (3) life prediction: and connecting the MHI and the THI to form a health index data set, and inputting the health index data set into a regression layer of the life prediction model of the aircraft engine to predict the residual service life.
2. The method of claim 1, wherein the model comprises a MCLSTM model, a statistical model, a full connection layer FC1, a full connection layer FC2, a full connection layer FC3, a regression layer, and the MCLSTM model comprises a hierarchical unit and a multi-cell update unit.
3. The method for predicting the life of the aero-engine based on the MCLSTM model as claimed in claim 2, wherein the specific step of extracting the health index MHI based on the MCLSTM model in the step 4) is as follows:
4-1-1) inputting the preprocessed data in the step 2) into an MCLSTM model as input data, performing hierarchical division on the input data through a hierarchical division unit, and updating the data after hierarchical division through a multi-cell updating unit to obtain hidden features;
4-1-2) inputting the obtained hidden features into an attention mechanism to obtain attention weights, and combining the attention weights with the hidden features; obtaining a merging characteristic;
4-1-3) inputting the merged characteristics into the full connection layer FC1 to obtain the output of the full connection layer FC1, and inputting the output of the full connection layer FC1 into the full connection layer FC2 to obtain the health index MHI based on the MCLSTM model.
4. The method for predicting the life of the aircraft engine based on the MCLSTM model as claimed in claim 2, wherein the method for extracting the health index THI based on the statistical model in the step 4) comprises the following steps:
inputting the preprocessed data in the step 2) into a statistical model, counting the mean value and the trend coefficient of the input data through the statistical model, inputting the statistical data into a full-connection layer FC3 of an aircraft engine service life prediction model, and obtaining a health index THI based on the statistical model.
5. The method for predicting the life of the aircraft engine based on the MCLSTM model as claimed in claim 3, wherein the step 4-1-1) of obtaining the hidden features comprises the following specific steps:
4-1-1-1) hierarchical partitioning: taking the preprocessed data in the step 2) as input data, and taking the input data x at the time tt＝[xt,1 xt,2 ... xt,n]TAnd recursive data h of the MCLSTM model at time t-1t-1＝[ht-1,1 ht-1,2 ... ht-1,m]TInputting the output A into a full connection layer of a multi-cell long-short term memory neural network MCLSTM model, activating the output through a tanh activation function to obtain the output A of the full connection layer, and regularizing the output A of the full connection layer through a SoftMax function to obtain the output A of a hierarchical division unit1：
A＝tanh(wxaxt+whaht-1+ba) (1)
A1＝softmax(A)＝[a1 a2 a3 a4 a5] (2)
In the formulae (1) and (2), wxaAnd whaRespectively input data weight and recursive data weight in the full link layer, baFor full link layer bias, a1,a2,a3,a4And a5Respectively representing five levels, high, medium, low and medium, A1The element position of the middle maximum element represents the level of data input into the MCLSTM at the time t, and n and m are the characteristic dimensions of the input data and the recursion data respectively;
4-1-1-2) extracting hidden features: inputting data incapable of hierarchy into the multi-cell update unit c according to the hierarchy division structuretFive different dissimilatory units ct(m)、ct(i)、ct(j)、ct(k)、ct(l) According to the update results of five different units and the hierarchy division result A1The output c of the multi-cell renewal unit is obtained by weighting and combining the differentiation outputs of the sub-cell units in the multi-cell renewal unittAnd extracting hidden feature h at time tt：
Figure FDA0003325770570000021
In the formula (3), ct(m) is high level data a1Corresponding data update unit, ct(l) For medium and high level data a2Corresponding data update unit, ct(k) Is middle level data a3Corresponding data update unit, ct(j) For medium and low level data a4Corresponding data update unit, ct(i) As low-level data a5Corresponding data update unit, sigma is sigmod activation function, wixAs a weight between the input data and the MCLSTM model input gate, wihFor the weight between the recursive data and the input gate of the MCLSTM model, wfxAs a weight between the input data and the MCLSTM model forget gate, wfhIs the weight between the recursive data and the MCLSTM model forget gate, woxAs a weight between the input data and the output gate of the MCLSTM model, wohAs weights between the recursive data and the output gates of the MCLSTM model, biInput gate bias for MCLSTM model, bfForget gate bias for MCLSTM model, boOutput gate offset for MCLSTM model, ct-1Indicates the memory cell output, w, at time t-1cxAs a weight between the input data and the cell memory unit, wchWeights between recursive data and cell memory units, bcBias for cell memory cell in MCLSTM model, <' > is dot product operation, s1、s2All are mixing scale coefficients obtained by learning.
CN202111261118.6A 2021-10-28 2021-10-28 Aero-engine service life prediction method based on MCLSTM model Pending CN113836822A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111261118.6A CN113836822A (en) 2021-10-28 2021-10-28 Aero-engine service life prediction method based on MCLSTM model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111261118.6A CN113836822A (en) 2021-10-28 2021-10-28 Aero-engine service life prediction method based on MCLSTM model

Publications (1)

Publication Number Publication Date
CN113836822A true CN113836822A (en) 2021-12-24

Family

ID=78966262

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111261118.6A Pending CN113836822A (en) 2021-10-28 2021-10-28 Aero-engine service life prediction method based on MCLSTM model

Country Status (1)

Country Link
CN (1) CN113836822A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115470717A (en) * 2022-10-31 2022-12-13 四川工程职业技术学院 Method, device, equipment and storage medium for predicting remaining life of robot
CN115713044A (en) * 2023-01-09 2023-02-24 佰聆数据股份有限公司 Method and device for analyzing residual service life of electromechanical equipment under multi-working-condition switching

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109472110A (en) * 2018-11-29 2019-03-15 南京航空航天大学 A kind of aero-engine remaining life prediction technique based on LSTM network and ARIMA model
CN110618671A (en) * 2018-06-20 2019-12-27 安波福技术有限公司 Over-the-air (OTA) mobile service platform
US20200019844A1 (en) * 2017-04-28 2020-01-16 Intel Corporation Smart autonomous machines utilizing cloud, error corrections, and predictions
CN111723527A (en) * 2020-06-28 2020-09-29 重庆大学 Gear residual life prediction method based on cocktail long-term and short-term memory neural network
CN113297787A (en) * 2021-05-12 2021-08-24 北京航空航天大学 Method for predicting remaining life of aircraft engine based on transfer learning
CN113505535A (en) * 2021-07-08 2021-10-15 重庆大学 Motor life prediction method based on gate control self-adaptive hierarchical attention unit network

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200019844A1 (en) * 2017-04-28 2020-01-16 Intel Corporation Smart autonomous machines utilizing cloud, error corrections, and predictions
CN110618671A (en) * 2018-06-20 2019-12-27 安波福技术有限公司 Over-the-air (OTA) mobile service platform
CN109472110A (en) * 2018-11-29 2019-03-15 南京航空航天大学 A kind of aero-engine remaining life prediction technique based on LSTM network and ARIMA model
CN111723527A (en) * 2020-06-28 2020-09-29 重庆大学 Gear residual life prediction method based on cocktail long-term and short-term memory neural network
CN113297787A (en) * 2021-05-12 2021-08-24 北京航空航天大学 Method for predicting remaining life of aircraft engine based on transfer learning
CN113505535A (en) * 2021-07-08 2021-10-15 重庆大学 Motor life prediction method based on gate control self-adaptive hierarchical attention unit network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
SHENG XIANG等: "Multicellular LSTM-based deep learning model for aero-engine remaining useful life prediction", pages 1 - 10, Retrieved from the Internet <URL:《网页在线公开：https://doi.org/10.1016/j.ress.2021.107927》> *
林鹏飞;陶继忠;: "基于多样性特征和多源信息的分子泵故障诊断", 真空科学与技术学报, no. 01, pages 37 - 43 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115470717A (en) * 2022-10-31 2022-12-13 四川工程职业技术学院 Method, device, equipment and storage medium for predicting remaining life of robot
CN115713044A (en) * 2023-01-09 2023-02-24 佰聆数据股份有限公司 Method and device for analyzing residual service life of electromechanical equipment under multi-working-condition switching

Similar Documents

Publication Publication Date Title
CN106446540B (en) A kind of Wind turbines health status real time evaluating method
Andrews et al. Learning high-dimensional directed acyclic graphs with mixed data-types
CN113836822A (en) Aero-engine service life prediction method based on MCLSTM model
Bai et al. A forecasting method of forest pests based on the rough set and PSO-BP neural network
CN114218872B (en) DBN-LSTM semi-supervised joint model-based residual service life prediction method
CN110990718A (en) Social network model building module of company image improving system
Wang et al. A remaining useful life prediction model based on hybrid long-short sequences for engines
CN116187835A (en) Data-driven-based method and system for estimating theoretical line loss interval of transformer area
Xu et al. Time series chain graph for modeling reliability covariates in degradation process
CN115269861A (en) Reinforced learning knowledge graph reasoning method based on generative confrontation and imitation learning
CN113706328A (en) Intelligent manufacturing capability maturity evaluation method based on FASSA-BP algorithm
CN113743453A (en) Population quantity prediction method based on random forest
CN113435101A (en) Power failure prediction method for support vector machine based on particle swarm optimization
CN113987704A (en) Aero-engine service life prediction method based on automatic multi-differential learning deep network
CN116341371A (en) Bi-LSTM-CORAL network migration learning-based method for predicting residual life of aeroengine
Dui et al. Reliability Evaluation and Prediction Method with Small Samples.
Chen Hotel management evaluation index system based on data mining and deep neural network
Yacoubi et al. A Multiobjective Crystal Optimization-based association rule mining enhanced with TOPSIS for predictive maintenance analysis
CN114091961A (en) Power enterprise supplier evaluation method based on semi-supervised SVM
Dong et al. Security situation prediction method for industrial control network based on adaptive Grey Verhulst model and GRU network
CN113868964A (en) Aero-engine service life prediction method based on time-space multi-differentiation processing deep network
Bewley et al. Reward Learning with Trees: Methods and Evaluation
Yuan et al. Early Detecting the At-risk Students in Online Courses Based on Their Behavior Sequences
CN112561203B (en) Method and system for realizing water level early warning based on clustering and GRU
Park et al. CRAN Task View: Bayesian Inference

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