CN115017819A - Engine remaining service life prediction method and device based on hybrid model - Google Patents

Engine remaining service life prediction method and device based on hybrid model Download PDF

Info

Publication number
CN115017819A
CN115017819A CN202210698982.0A CN202210698982A CN115017819A CN 115017819 A CN115017819 A CN 115017819A CN 202210698982 A CN202210698982 A CN 202210698982A CN 115017819 A CN115017819 A CN 115017819A
Authority
CN
China
Prior art keywords
model
sample
engine
life
predicted
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
CN202210698982.0A
Other languages
Chinese (zh)
Inventor
郭钧
李大鹏
杜百岗
彭兆
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.)
Wuhan University of Technology WUT
Original Assignee
Wuhan University of Technology WUT
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 Wuhan University of Technology WUT filed Critical Wuhan University of Technology WUT
Priority to CN202210698982.0A priority Critical patent/CN115017819A/en
Publication of CN115017819A publication Critical patent/CN115017819A/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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Geometry (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The application discloses a method and a device for predicting the remaining service life of an engine based on a hybrid model, wherein the method comprises the following steps: obtaining a plurality of groups of sensor data samples of a sample engine; establishing an initial mixed model, and obtaining a corresponding first predicted residual life sample/second predicted residual life sample through stacking integration; inputting the first predicted residual life sample and the second predicted residual life sample into an initial linear regression model, and iteratively training the linear regression model; and determining the residual service life of the engine to be tested according to the data of the multiple groups of sensors of the engine to be tested by using the target mixed model and the linear regression model which are trained completely. The reliability of the obtained first predicted residual life and the second predicted residual life is high through a stacking integration technology; in addition, the linear regression model is used for carrying out weight redistribution on the first predicted residual life and the second predicted residual life, so that the dependence on a certain prediction result is reduced, and the precision of the prediction result is improved.

Description

Engine residual service life prediction method and device based on hybrid model
Technical Field
The invention relates to the field of aero-engine fault prediction and health management, in particular to a method and a device for predicting the remaining service life of an engine based on a hybrid model.
Background
As manufacturing industries and industrial systems mature, various mechanical equipment and devices play a very important role for each industry. Aircraft engines are critical components of aircraft, and if faults occur suddenly during operation, a huge disaster is caused. The Predictive Health Management (PHM) of a device is a good management solution for the device. Prediction of the remaining service life of an aircraft engine is an important one of the PHM technologies.
In order to obtain the remaining service life of the device, the commonly used remaining service life prediction methods are roughly classified into three types: traditional physical model based methods; a data-driven method; two methods are mixed. Data-driven methods are often used because methods based on physical models are difficult to give accurate predictions and mixed methods are difficult to implement. However, the existing data-driven method mainly uses a deep learning model to learn and process sensor data so as to obtain a prediction result, and the prediction precision is limited.
Therefore, the prior art has the problem of low prediction precision of the residual service life of the engine.
Disclosure of Invention
In view of the above, it is necessary to provide a method and a device for predicting remaining useful life of an engine based on a hybrid model, so as to solve the problem of low accuracy of predicting remaining useful life of an engine in the prior art.
In order to solve the above problems, the present invention provides a method for predicting remaining useful life of an engine based on a hybrid model, comprising:
acquiring a plurality of groups of sensor data samples of a sample engine;
establishing an initial mixed model, wherein the initial mixed model comprises a first model and a second model; respectively inputting a plurality of groups of sensor data samples into a first model/a second model, taking a first residual life sample/a second residual life sample as output, obtaining a first predicted residual life sample/a second predicted residual life sample of the plurality of groups of sensor data samples through stacking integration, iteratively training to a first preset number of times, and determining a target mixed model which is completely trained;
inputting the first predicted residual life sample and the second predicted residual life sample into an initial linear regression model, and carrying out iterative training to a second preset number of times by taking the residual service life of the sample engine as output to obtain a target linear regression model with complete training;
and acquiring multiple groups of sensor data of the engine to be tested, and determining the residual service life of the engine to be tested based on the target mixed model and the target linear regression model.
Further, respectively inputting multiple groups of sensor data samples into the first model/the second model, obtaining first predicted residual life samples/second predicted residual life samples of the multiple groups of sensor data samples by stacking integration by using the first residual life samples/the second residual life samples as output, iteratively training to a first preset number of times, and determining a target mixed model with complete training, wherein the method comprises the following steps of:
dividing a plurality of groups of sensor data samples into K parts, respectively inputting the K parts into a first model and a second model, wherein K-1 part takes a corresponding first residual life sample/second residual life sample as output, the rest takes an initial first predicted residual life sample/initial second predicted residual life sample as output, K-fold cross validation is carried out through stacking integration, iterative calculation is carried out to a first preset number of times, and the initial first predicted residual life sample/initial second predicted residual life sample corresponding to the plurality of groups of sensor data samples are determined;
first predicted residual life samples and second predicted residual life samples of the multiple sets of sensor data samples are determined by an averaging operation based on the initial first predicted residual life samples/the initial second predicted residual life samples.
Further, the first model comprises a first input layer, a first attention mechanism module, a time convolution neural network module, a first full connection layer and a first output layer which are connected in sequence.
Further, the second model comprises a second input layer, a convolutional neural network module, a bidirectional gated cyclic neural network module, a second attention mechanism module, a second full-link layer and a second output layer which are connected in sequence.
Further, obtaining a plurality of sets of sensor data samples for a sample engine, comprising:
acquiring multiple groups of sensor data initial samples of a sample engine, and removing abnormal values to obtain multiple groups of sensor characteristic samples;
and carrying out normalization processing on the multiple groups of sensor characteristic samples to obtain multiple groups of processed sensor data samples.
Further, obtaining a first predicted remaining life sample/a second predicted remaining life sample of the plurality of sets of sensor data samples by stacked integration includes:
and processing the first predicted residual life initial sample/the second predicted residual life initial sample through a segmentation line function to obtain a first predicted residual life sample/a second predicted residual life sample of the multiple groups of sensor data samples.
Further, obtaining a well-trained target linear regression model, further comprising:
establishing an evaluation index, and evaluating a test result of the linear regression model;
and adjusting the relevant parameters of the linear regression model according to the evaluation result to obtain a target linear regression model with complete training.
In order to solve the above problem, the present invention further provides a device for predicting remaining useful life of an engine based on a hybrid model, comprising:
the system comprises a sample acquisition module, a data acquisition module and a data processing module, wherein the sample acquisition module is used for acquiring a plurality of groups of sensor data samples of a sample engine;
the hybrid model acquisition module is used for establishing an initial hybrid model, wherein the initial hybrid model comprises a first model and a second model; respectively inputting a plurality of groups of sensor data samples into a first model/a second model, taking a first residual life sample/a second residual life sample as output, obtaining a first predicted residual life sample/a second predicted residual life sample of the plurality of groups of sensor data samples through stacking integration, iteratively training to a first preset number of times, and determining a target mixed model which is completely trained;
the linear regression model acquisition module is used for inputting the first predicted residual life sample and the second predicted residual life sample into the initial linear regression model, iteratively training to a second preset number of times by taking the residual service life of the sample engine as output, and obtaining a target linear regression model with complete training;
and the residual service life determining module is used for acquiring multiple groups of sensor data of the engine to be tested and determining the residual service life of the engine to be tested based on the target mixed model and the target linear regression model.
In order to solve the above problem, the present invention also provides a storage medium storing computer program instructions which, when executed by a computer, cause the computer to execute the hybrid model-based engine remaining useful life prediction method as described above.
The beneficial effect of adopting above-mentioned technical scheme is: the invention provides a method and a device for predicting the remaining service life of an engine based on a hybrid model, wherein the method comprises the following steps: respectively training a first model and a second model in the initial mixed model according to a sensor data sample of a sample engine, and determining a target mixed model which is completely trained; then, obtaining a first predicted residual life and a second predicted residual life of a sensor data sample of the sample engine through a stacking integration technology; then, respectively inputting the first predicted residual life and the second predicted residual life into an initial linear regression model, and obtaining a target linear regression model with complete training through training; and finally, predicting the residual service life of the engine to be tested based on the target mixed model and the target linear regression model. According to the method, the hybrid model composed of the multiple models is arranged, the residual service life of the engine is predicted by the aid of the stacking integration technology, the residual service life of the engine is predicted from multiple angles and multiple directions, reliability of the residual service life of the engine is improved, then the multiple prediction results in the hybrid model are organically combined by the linear regression model, weight redistribution is carried out on the prediction result of each single model, dependence on a certain prediction result is reduced, and accuracy of the prediction result is improved.
Drawings
FIG. 1 is a schematic flow chart illustrating an embodiment of a method for predicting remaining useful life of an engine based on a hybrid model according to the present invention;
FIG. 2 is a schematic structural diagram of a first model according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a second embodiment of a model according to the present invention;
FIG. 4 is a schematic structural diagram of a hybrid model-based engine remaining service life prediction apparatus provided in the present invention;
fig. 5 is a block diagram of an embodiment of an electronic device provided in the present invention.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
Before the embodiments are set forth, attention mechanisms, convolutional neural networks, time-convolutional neural networks, cyclic neural networks are described:
the attention mechanism is one that enables a neural network to be provided with focus on a subset of its inputs (or features), enabling selection of a particular input. Attention can be applied to any type of input, and in the case of limited computing power, an attention mechanism (attention mechanism) is a resource allocation scheme that is the main means for solving the information overload problem, and computing resources are allocated to more important tasks.
Among them, attention is generally classified into two types: one is conscious attention from top to bottom, called focused (focus) attention, which refers to attention that has a predetermined purpose, is task-dependent, and is actively and consciously focused on a certain subject; the other is a bottom-up unconscious attention, called saliency-based (saliency-based) attention, which is attention driven by external stimuli, does not require active intervention, and is also task-independent. If a subject's stimulation information differs from its surrounding information, an unconscious "winner-take-all" or gating (gating) mechanism may divert attention to the subject. Regardless of whether such attention is intended or unintended, most human brain activities require attention, such as memorizing information, reading or thinking, and the like.
The Convolutional Neural Network (CNN) is a deep neural network with a convolutional structure, the convolutional structure can reduce the memory occupied by the deep network, and three key operations are that the convolutional neural network is a local receptive field, the convolutional neural network is weight sharing, and the convolutional neural network is a posing layer, so that the parameter number of the neural network can be effectively reduced, and the overfitting problem of a model is relieved.
A time Convolutional neural network (Temporal Convolutional Networks) starts with a general architecture of Convolutional sequence prediction, not an architecture, but a type of architecture. Characteristics of TCN networks include: 1. the relationship in convolution is causal, meaning that there is no information never coming into the past. 2. This architecture can take an arbitrarily long sequence as input and map to the same length output. 3. Residual layer and hole convolution techniques are used.
A Recurrent Neural Network (RNN) is a type of Recurrent Neural Network (Recurrent Neural Network) in which sequence data is input, recursion is performed in the direction of evolution of the sequence, and all nodes (Recurrent units) are connected in a chain.
In the gradient calculation method in the recurrent neural network, when the number of time steps is large or the time steps is small, the gradient of the recurrent neural network is liable to be attenuated or exploded. Therefore, it is difficult for the recurrent neural network to capture the dependence of the time step distance in the time series in practice.
The proposal of the gated recurrent neural network (gated recurrent neural network) can better capture the dependence relationship with larger time step distance in the time sequence. It controls the flow of information through the door that can be learned. Among them, a Gated Recurrent Unit (GRU) is a commonly used gated recurrent neural network.
At present, in order to predict the remaining service life of an engine, a deep learning mode is generally adopted to train a neural network model, and then the well-trained neural network model is used to predict the remaining service life of the engine.
However, because the structure of each single deep learning model is different and the data characteristics focused on are different, the prediction precision in the prior art is limited, and the prediction precision of the residual service life of the engine is low.
In order to solve the problem that the single deep learning neural network has low precision when predicting the remaining service life of the engine in the prior art, the invention provides a method and a device for predicting the remaining service life of the engine based on a hybrid model, an electronic device and a storage medium, which are respectively described in detail below.
As shown in fig. 1, fig. 1 is a schematic flowchart of an embodiment of a method for predicting remaining useful life of an engine based on a hybrid model provided by the present invention, including:
step S101: multiple sets of sensor data samples of a sample engine are obtained.
Step S102: establishing an initial mixed model, wherein the initial mixed model comprises a first model and a second model; respectively inputting a plurality of groups of sensor data samples into a first model/a second model, taking a first residual life sample/a second residual life sample as output, obtaining a first predicted residual life sample/a second predicted residual life sample of the plurality of groups of sensor data samples through stacking integration, iteratively training to a first preset number of times, and determining a target mixed model which is completely trained.
Step S103: and inputting the first predicted residual life sample and the second predicted residual life sample into the initial linear regression model, and performing iterative training to a second preset time by taking the residual service life of the sample engine as output to obtain a target linear regression model with complete training.
Step S104: and acquiring multiple groups of sensor data of the engine to be tested, and determining the residual service life of the engine to be tested based on the target mixed model and the target linear regression model.
In this embodiment, first, a plurality of groups of sensor data samples of the sample engine are obtained, and in order to improve the training level of the model, the number of the sensor data samples should be as large as possible; secondly, establishing an initial mixed model, wherein the initial mixed model comprises a first model and a second model; respectively inputting multiple groups of sensor data samples into a first model/a second model, training the first model and the second model by taking a first residual life sample/a second predicted life sample as output, obtaining a first predicted residual life sample and a second predicted residual life sample corresponding to the multiple groups of sensor data samples through stacking integration, and performing iterative training to a first preset number of times to obtain a first model and a second model which are completely trained, thereby determining a target mixed model which is completely trained; then, inputting the first predicted residual life sample and the second predicted residual life sample into an initial linear regression model, taking the residual service life of the sample engine as output, and performing iterative training to a second preset number of times to obtain a target linear regression model with complete training; finally, in order to determine the residual service life of the engine to be tested, inputting a plurality of groups of sensor data of the engine to be tested into a target mixed model which is completely trained, and respectively obtaining a first predicted residual life and a second predicted residual life; and then, respectively inputting the first predicted residual life and the second predicted residual life into a target linear regression model, so that the residual service life of the engine to be tested can be determined.
It can be understood that, in the above embodiment, the stacking integration technology is used, and the corresponding first predicted remaining life and second predicted remaining life are obtained according to multiple sets of sensor data of the engine to be tested, because the stacking integration technology traverses all sensor data in the using process, and iterates and calculates for multiple times, and also averages the results of multiple times of prediction, the first predicted remaining life and the second predicted remaining life obtained by using the stacking integration have higher accuracy; further, the final residual service life of the engine to be tested is determined according to the first predicted residual service life and the second predicted residual service life by using the target linear regression model, so that the prediction effects of the first model and the second model are effectively utilized, the prediction effects of the first model and the second model are organically distributed and optimized, and the precision of the finally obtained residual service life of the engine to be tested is improved.
In summary, in order to improve the accuracy of the prediction result, the reliability of the obtained first predicted remaining life/second predicted remaining life is high based on the stacked integration technology in the embodiment; in addition, the weight of the prediction results of the two models is redistributed by using the target linear regression model, so that the dependence on a certain prediction result is reduced, and the precision of the prediction result is improved.
As a preferred embodiment, in step S101, a plurality of sets of sensor data samples of a sample engine are acquired, including:
firstly, acquiring historical failure data of a sample engine, and removing abnormal values to obtain a sensor characteristic change sample; and then, carrying out normalization processing on the sensor characteristic change samples to obtain a plurality of groups of sensor data samples.
In a specific embodiment, historical failure data of a sample engine is obtained, all sensor data are listed, and sensor data which are not changed are removed to obtain a sensor characteristic change sample; that is, the sensor data which is not changed in the historical failure data is removed, and a sensor characteristic change sample is obtained.
Further, normalization processing needs to be performed on the sensor characteristic change samples, so as to obtain sensor data samples. In this embodiment, a Min-Max normalization method is adopted to scale the feature data to the range of [0,1], and the specific formula is as follows:
Figure BDA0003703756270000091
wherein x normal Is normalized data, x i Is the value in the ith signature sequence, minx i Is the minimum value in the signature sequence, maxx i Is in the feature sequenceA maximum value.
In the embodiment, by performing normalization processing on the sensor characteristic change sample, not only can adverse effects caused by singular sample data be eliminated, but also the difficulty of calculation can be reduced, and the prediction precision can be improved.
As a preferred embodiment, in step S102, after the multiple sets of sensor data samples are respectively input into the first model and the second model, the first predicted remaining life sample and the second predicted remaining life sample are determined by performing K-fold cross validation through stacking integration.
Dividing a plurality of groups of sensor data samples into K parts, respectively inputting the K parts into a first model/a second model, wherein K-1 part takes a corresponding first residual life sample/second residual life sample as output, the rest takes an initial first predicted residual life sample/initial second predicted residual life sample as output, K-fold cross validation is carried out through stacking integration, iterative calculation is carried out to a first preset number of times, and the initial first predicted residual life sample/initial second predicted residual life sample corresponding to the plurality of groups of sensor data samples are determined;
first predicted remaining life samples and second predicted remaining life samples of the plurality of sets of sensor data samples are determined by an averaging operation based on the initial first predicted remaining life samples/the initial second predicted remaining life samples.
In a specific embodiment, the value of K may be 5, and in other embodiments, K may theoretically take any non-zero positive integer.
Further, after the first predicted remaining life sample and the second predicted remaining life sample with high reliability are obtained, since the remaining service life of the engine has a threshold interval, the first predicted remaining life sample and the second predicted remaining life sample need to be corrected in order to meet practical problems.
In one embodiment, the predicted remaining life sample is modified by a piecewise function having a maximum remaining useful life of 130, the piecewise linear function formula being as follows:
Figure BDA0003703756270000101
wherein RUL is a modified sample of predicted remaining life, RUL real Is the true value of the predicted remaining life sample, and 130 is the maximum value.
Preferably, the first model includes a first attention mechanism, a time-convolutional neural network, and a first fully-connected layer. As shown in fig. 2, fig. 2 is a schematic structural diagram of an embodiment of a first model provided by the present invention, and the first model 200 includes a first input layer 201, a first attention mechanism module 202, a time convolution neural network module 203, a first fully-connected layer 204, and a first output layer 205, which are connected in sequence.
Further, the time convolutional neural network module 203 includes an extended causal convolution, weight initialization, a Relu activation layer, a discard layer, and a convolutional layer, which is a 1 × 1 convolutional layer; the first fully-connected layer 204 includes two fully-connected layers.
In a specific embodiment, a plurality of groups of sensor data samples are input to the first input layer 201, and are processed by the first attention mechanism module 202, so as to increase attention to the characteristics of the important sensor data samples, thereby improving the robustness of the first model 200; secondly, the time convolution neural network module 203 mainly comprises a residual error network and an extended causal convolution, the sensor data sample is subjected to feature extraction through the extended causal convolution, the residual error module increases the receptive field of the model and improves the learning capability of the model, the sensor features of the sensor data sample are extracted by learning the sensor features in the sensor data sample, so that the first model 200 can better learn the dependence relationship between the sensor features and the residual service life of the engine, wherein the activation function of the time convolution neural network module 203 is a relu function; then, mapping the learning result through the first fully-connected layer 204, namely mapping the predicted value of the remaining service life of the sample engine, wherein the activation function of the first fully-connected layer 204 of the first layer is a relu function, and the activation function of the first fully-connected layer 204 of the second layer is a linear function; finally, a predicted value of the remaining useful life of the sample engine is output from the first output layer 205.
In one embodiment, the calculation formula of the time convolution neural network module 203 is:
Figure BDA0003703756270000111
wherein the content of the first and second substances,
Figure BDA0003703756270000112
is a one-dimensional sequence of inputs that,
Figure BDA0003703756270000113
is a filter, d is the expansion factor, k is the size of the filter, and s-d · i illustrates the past direction.
Preferably, the second model includes a convolutional neural network, a bi-directional gated cyclic neural network, a second attention mechanism, and a second fully-connected layer. As shown in fig. 3, fig. 3 is a schematic structural diagram of an embodiment of a second model provided by the present invention, and the second model 300 includes a second input layer 301, a convolutional neural network module 302, a bidirectional gated recurrent neural network module 303, a second attention mechanism module 304, a second fully connected layer 305, and a second output layer 306, which are connected in sequence.
Further, the convolutional neural network module 302 includes convolutional layers and pooling layers; the bidirectional gated recurrent neural network module 303 includes a forward BLSTM and a reverse BLSTM, and the second fully-connected layer 305 includes two fully-connected layers.
In one embodiment, a plurality of sets of sensor data samples are input to the second input layer 301, and first, the features of the sensors are extracted through the convolutional layer and the pooling layer of the convolutional neural network module 302; secondly, the bidirectional gated recurrent neural network module 303 can quickly learn the relationship between the characteristics of the sensor and the tags based on the characteristics of the sensor, and determine the time dependence of the related sequence of the sensor data; then, the characteristics more relevant to predicting the remaining service life of the engine are focused through the processing of the second attention mechanism module 304; finally, mapping the learning result, namely the predicted value of the remaining service life of the sample engine, through the second full connection layer 305; finally, a predicted value of the remaining useful life of the sample engine is output by the second output layer 306.
The activation function adopted by the convolutional neural network module 302 is a tanh function, the activation function adopted by the bidirectional gated recurrent neural network module 303 is a relu function, the activation function of the first layer second fully-connected layer 305 is a relu function, and the activation function of the second layer second fully-connected layer 305 is a linear function.
In this embodiment, based on the convolutional neural network module 302, the sensor data characteristics of the sample engine can be quickly extracted, based on the bidirectional gated cyclic neural network module 303, the quick convergence can be realized, and based on the second attention mechanism module 304, the robustness of the prediction result can be improved, so that the second model 300 can obtain a better prediction result of the remaining service life of the sample engine.
In one embodiment, the calculation formula of the convolutional neural network module 302 is:
Figure BDA0003703756270000121
where f is the feature generated, u is the feature of the input, k represents the convolution kernel, b is the bias term,
Figure BDA0003703756270000131
is an activation function.
In one embodiment, the calculation formula of the bidirectional gated recurrent neural network module 303 is:
Figure BDA0003703756270000132
Figure BDA0003703756270000133
Figure BDA0003703756270000134
where G () is an internal network function, w t ,v t Is the weight of the forward BLSTM output and the backward BLSTM output at time t, b t Is the deviation of the output layer, h t Representing the cell state at time t, x t Which represents the input at the time t,
Figure BDA0003703756270000135
representing the forward cell state at time t-1,
Figure BDA0003703756270000136
indicating the backward cell state at time t-1.
In one embodiment, the calculation formula of the second attention mechanism module 304 is:
a t ＝tanh(W a h t +b)
Figure BDA0003703756270000137
Figure BDA0003703756270000138
wherein, W a Is the weight matrix, b is the bias term, h t Represents an input, a t Representing the normalized result of the input, c t Representing a mechanism of attention.
After the first model and the second model which are trained completely are obtained, the first predicted residual life sample and the second predicted residual life sample can be corrected. In general, the first predicted remaining life sample and the second predicted remaining life sample are not equal because the first model and the second model are not identical.
In other embodiments, the hybrid model may also include other numbers of multiple other types of models, and the remaining life of the engine is predicted at the same time, so as to obtain multiple predicted results.
Further, in order to effectively utilize the first model and the second model, unify the remaining life prediction results of the engine, and improve the accuracy of the prediction results, a Linear Regression (Linear Regression) model is also used as a learner in the application, the first predicted remaining life sample and the second predicted remaining life sample are input into the initial Linear Regression model, the remaining service life of the engine of the sample is used as output, and the Linear Regression model is iteratively trained to a second preset number of times, so that a target Linear Regression model with complete training is obtained.
In order to check the precision of the prediction result according to the linear regression model, an evaluation index is also established to evaluate the prediction result, wherein the evaluation index is as follows:
Figure BDA0003703756270000141
Figure BDA0003703756270000142
where n represents the total number of sample engines,
Figure BDA0003703756270000143
representing the error between the predicted value and the true value of the ith sample engine.
Finally, in order to clearly explain the process of obtaining the remaining service life of the engine to be tested according to the multiple sets of sensor data of the engine to be tested through the well-trained target hybrid model and the linear regression model in detail, the following specific embodiments are provided.
The description is given in this embodiment by selecting the C-MPASS data set provided by NASA with respect to an aircraft engine. The C-MPASS dataset is a turbofan engine dataset that is widely used for RUL prediction. It has four sub data sets (FD001 to FD004), each of which has historical data of the turbofan engine under different operating conditions and failure modes. The present embodiment uses a first sub data set (FD001) that has degradation data for 100 engines.
First, the changes in the sensor data in the data set are listed, including 21 sensor data and 3 operating parameters, no change in the sensor data for numbers 1, 5, 10, 16, 18, 19, and no change in the third operating parameter, so these parameters are discarded, taking the current engine operating cycle as one of the selected characteristics.
Secondly, scaling the characteristic data to be in a [0,1] range by adopting a Min-Max normalization method; and (4) correcting the residual service life of the engine by adopting a piecewise linear function, and setting a piecewise linear function threshold value to be 130.
Then, according to the running period of the engine to be measured, the sensor data is cut to construct sliding time windows with different sizes, in the embodiment, 30, 60, 90 and 120 are selected as cutting standards of the time windows, and the sensor data is divided into different data sets.
And then, respectively inputting the data sets into the first model and the second model to respectively obtain a first predicted residual life and a second predicted residual life, and performing K-fold cross validation on the first model and the second model by stacking an integration frame in the integrated learning to obtain a new training set and a new test set.
And finally, the linear regression model is used as a final learner, and the obtained new training set and the test set are input into the final learner to obtain the residual service life of the engine to be tested.
Further, the effect of the method can be evaluated by comparing the prediction accuracy of the method with that of other methods, and the results of the RMSE and Score evaluation of the method with those of other methods are shown in the following table:
Figure BDA0003703756270000151
Figure BDA0003703756270000161
in conclusion, the method for predicting the remaining service life of the engine based on the hybrid model can accurately predict the remaining service life of the engine. As can be seen from the above table, the performance index of the method is far better than that of other methods. According to the invention, a plurality of models are constructed to learn data, and then the results after model learning are relearned, so that the accuracy of the prediction result is further improved.
In order to solve the above problem, the present invention further provides a hybrid model-based engine remaining useful life prediction apparatus, as shown in fig. 4, where fig. 4 is a schematic structural diagram of the hybrid model-based engine remaining useful life prediction apparatus provided in the present invention, and the hybrid model-based engine remaining useful life prediction apparatus 400 includes:
a sample acquisition module 401 for acquiring a plurality of sets of sensor data samples of a sample engine;
a hybrid model obtaining module 402, configured to establish an initial hybrid model, where the initial hybrid model includes a first model and a second model; respectively inputting multiple groups of sensor data samples into a first model/a second model, taking a first residual life sample/a second residual life sample as output, obtaining a first predicted residual life sample/a second predicted residual life sample of the multiple groups of sensor data samples through stacking integration, iteratively training to a first preset number of times, and determining a target mixed model with complete training;
the linear regression model obtaining module 403 is configured to input the first predicted remaining life sample and the second predicted remaining life sample to the initial linear regression model, perform iterative training to a second preset number of times with the remaining service life of the sample engine as output, and obtain a target linear regression model with complete training;
and a remaining service life determining module 404, configured to obtain multiple sets of sensor data of the engine to be tested, and determine a remaining service life of the engine to be tested based on the target hybrid model and the target linear regression model.
The present invention further provides an electronic device, as shown in fig. 5, fig. 5 is a block diagram of an embodiment of the electronic device provided in the present invention. The electronic device 500 may be a computing device such as a mobile terminal, a desktop computer, a notebook, a palmtop, and a server. The electronic device 500 includes a processor 501 and a memory 502, wherein the memory 502 has stored thereon a hybrid model-based engine remaining useful life prediction program 503.
The memory 502 may be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device, in some embodiments. The memory 502 may also be an external storage device of the computer device in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), etc. provided on the computer device. Further, the memory 502 may also include both internal and external storage units of the computer device. The memory 502 is used for storing application software installed on the computer device and various data, such as program codes for installing the computer device. The memory 502 may also be used to temporarily store data that has been output or is to be output. In one embodiment, the hybrid model-based engine remaining useful life prediction program 503 may be executed by the processor 501 to implement the hybrid model-based engine remaining useful life prediction method according to various embodiments of the present invention.
Processor 501, which in some embodiments may be a Central Processing Unit (CPU), microprocessor or other data Processing chip, operates program code stored in memory 502 or processes data, such as executing a hybrid model-based engine remaining life prediction program.
The embodiment also provides a computer readable storage medium, on which a hybrid model-based engine remaining useful life prediction program is stored, and when the program is executed by a processor, the method for predicting the remaining useful life of an engine based on a hybrid model according to any one of the above technical solutions is implemented.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.

Claims (10)

1. A method for predicting the remaining service life of an engine based on a hybrid model is characterized by comprising the following steps:
obtaining a plurality of groups of sensor data samples of a sample engine;
establishing an initial mixed model, wherein the initial mixed model comprises a first model and a second model; respectively inputting the multiple groups of sensor data samples into the first model/the second model, taking a first residual life sample/a second residual life sample as output, obtaining a first predicted residual life sample/a second predicted residual life sample of the multiple groups of sensor data samples through stacking integration, iteratively training to a first preset number of times, and determining a target mixed model which is completely trained;
inputting the first predicted residual life sample and the second predicted residual life sample into an initial linear regression model, and performing iterative training to a second preset number of times by taking the residual service life of the sample engine as output to obtain a target linear regression model with complete training;
and acquiring multiple groups of sensor data of the engine to be tested, and determining the residual service life of the engine to be tested based on the target mixed model and the target linear regression model.
2. The method for predicting the remaining service life of the engine based on the hybrid model according to claim 1, wherein the inputting the plurality of groups of sensor data samples into the first model/the second model respectively, obtaining the first predicted remaining life sample/the second predicted remaining life sample of the plurality of groups of sensor data samples through stacking integration with the first remaining life sample/the second remaining life sample as output, iteratively training the samples to a first preset number of times, and determining a target hybrid model which is trained completely, comprises:
dividing the multiple groups of sensor data samples into K parts, and inputting the K parts into the first model and the second model respectively, wherein K-1 part takes a corresponding first residual life sample/second residual life sample as output, the rest takes an initial first predicted residual life sample/initial second predicted residual life sample as output, K-fold cross validation is carried out through stacking integration, iterative calculation is carried out to a first preset number of times, and the initial first predicted residual life sample/the initial second predicted residual life sample corresponding to the multiple groups of sensor data samples are determined;
determining a first predicted remaining life sample and a second predicted remaining life sample of the plurality of sets of sensor data samples by an averaging operation based on the initial first predicted remaining life sample/the initial second predicted remaining life sample.
3. The hybrid model-based engine remaining service life prediction method of claim 1, wherein the first model comprises a first input layer, a first attention mechanism module, a time convolution neural network module, a first fully connected layer, and a first output layer connected in sequence.
4. The hybrid model-based engine remaining service life prediction method of claim 1, wherein the second model comprises a second input layer, a convolutional neural network module, a bidirectional gated cyclic neural network module, a second attention mechanism module, a second fully connected layer, and a second output layer connected in sequence.
5. The hybrid model-based engine remaining useful life prediction method of claim 1, wherein said obtaining a plurality of sets of sensor data samples of a sample engine comprises:
acquiring multiple groups of sensor data initial samples of a sample engine, and removing abnormal values to obtain multiple groups of sensor characteristic samples;
and carrying out normalization processing on the multiple groups of sensor characteristic samples to obtain multiple groups of processed sensor data samples.
6. The hybrid model-based engine remaining life prediction method of claim 1, wherein said obtaining first/second predicted remaining life samples of said plurality of sets of sensor data samples by stacked integration comprises:
and processing the first predicted residual life initial sample/the second predicted residual life initial sample through a segmentation line function to obtain a first predicted residual life sample/a second predicted residual life sample of the multiple groups of sensor data samples.
7. The hybrid model-based engine remaining service life prediction method of claim 1, wherein the obtaining a well-trained target linear regression model further comprises:
establishing an evaluation index, and evaluating the test result of the linear regression model;
and adjusting the relevant parameters of the linear regression model according to the evaluation result to obtain a target linear regression model with complete training.
8. An engine remaining service life prediction apparatus based on a hybrid model, comprising:
the system comprises a sample acquisition module, a data acquisition module and a data processing module, wherein the sample acquisition module is used for acquiring a plurality of groups of sensor data samples of a sample engine;
the hybrid model acquisition module is used for establishing an initial hybrid model, wherein the initial hybrid model comprises a first model and a second model; respectively inputting the multiple groups of sensor data samples into the first model/the second model, obtaining first predicted residual life samples/second predicted residual life samples of the multiple groups of sensor data samples by stacking integration by taking the first residual life samples/the second residual life samples as output, iteratively training to a first preset number of times, and determining a target mixed model with complete training;
the linear regression model acquisition module is used for inputting the first predicted residual life sample and the second predicted residual life sample into an initial linear regression model, iteratively training to a second preset number of times by taking the residual service life of the sample engine as output, and obtaining a target linear regression model with complete training;
and the residual service life determining module is used for acquiring multiple groups of sensor data of the engine to be tested and determining the residual service life of the engine to be tested based on the target mixed model and the target linear regression model.
9. An electronic device comprising a processor and a memory, the memory having stored thereon a computer program which, when executed by the processor, carries out a hybrid model-based engine remaining useful life prediction method according to any one of claims 1-7.
10. A storage medium storing computer program instructions which, when executed by a computer, cause the computer to perform the hybrid model-based engine remaining useful life prediction method according to any one of claims 1 to 7.
CN202210698982.0A 2022-06-20 2022-06-20 Engine remaining service life prediction method and device based on hybrid model Pending CN115017819A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210698982.0A CN115017819A (en) 2022-06-20 2022-06-20 Engine remaining service life prediction method and device based on hybrid model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210698982.0A CN115017819A (en) 2022-06-20 2022-06-20 Engine remaining service life prediction method and device based on hybrid model

Publications (1)

Publication Number Publication Date
CN115017819A true CN115017819A (en) 2022-09-06

Family

ID=83076423

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210698982.0A Pending CN115017819A (en) 2022-06-20 2022-06-20 Engine remaining service life prediction method and device based on hybrid model

Country Status (1)

Country Link
CN (1) CN115017819A (en)

Similar Documents

Publication Publication Date Title
CN112131673B (en) Engine surge fault prediction system and method based on fusion neural network model
CN111210024B (en) Model training method, device, computer equipment and storage medium
US20210295162A1 (en) Neural network model training method and apparatus, computer device, and storage medium
CN112598248B (en) Load prediction method, load prediction device, computer equipment and storage medium
CN111625516A (en) Method and device for detecting data state, computer equipment and storage medium
CN111881023B (en) Software aging prediction method and device based on multi-model comparison
CN112433896B (en) Method, device, equipment and storage medium for predicting server disk faults
CN115656673A (en) Transformer data processing device and equipment storage medium
CN114943372A (en) Method and device for predicting life of proton exchange membrane based on Bayesian recurrent neural network
CN116126945A (en) Sensor running state analysis method and system based on data analysis
Padberg et al. Using machine learning for estimating the defect content after an inspection
CN113110961B (en) Equipment abnormality detection method and device, computer equipment and readable storage medium
Mohamed et al. Enhancing test cases prioritization for internet of things based systems using Search-based technique
CN116610983B (en) Abnormality analysis method and system for air purification control system
CN116665798A (en) Air pollution trend early warning method and related device
CN111783883A (en) Abnormal data detection method and device
CN116562120A (en) RVE-based turbine engine system health condition assessment method and RVE-based turbine engine system health condition assessment device
CN115017819A (en) Engine remaining service life prediction method and device based on hybrid model
US20220269991A1 (en) Evaluating reliability of artificial intelligence
CN113159100B (en) Circuit fault diagnosis method, circuit fault diagnosis device, electronic equipment and storage medium
CN116307183A (en) Engine residual service life prediction method and device, electronic equipment and medium
CN114417495A (en) Uncertainty-based satellite component layout reliability analysis method
CN117805637A (en) Battery safety monitoring method and system
CN118034733A (en) Automobile fault detection model updating method, device, equipment and storage medium
CN117150392A (en) Power load real-time classification decomposition method and system based on deep learning

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