CN116307183A - Engine residual service life prediction method and device, electronic equipment and medium - Google Patents
Engine residual service life prediction method and device, electronic equipment and medium Download PDFInfo
- Publication number
- CN116307183A CN116307183A CN202310270179.1A CN202310270179A CN116307183A CN 116307183 A CN116307183 A CN 116307183A CN 202310270179 A CN202310270179 A CN 202310270179A CN 116307183 A CN116307183 A CN 116307183A
- Authority
- CN
- China
- Prior art keywords
- engine
- service life
- sample
- data
- prediction model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 68
- 230000015556 catabolic process Effects 0.000 claims abstract description 117
- 238000006731 degradation reaction Methods 0.000 claims abstract description 117
- 230000007246 mechanism Effects 0.000 claims abstract description 37
- 230000004927 fusion Effects 0.000 claims abstract description 32
- 238000012545 processing Methods 0.000 claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 30
- 230000000873 masking effect Effects 0.000 claims abstract description 15
- 239000010410 layer Substances 0.000 claims description 85
- 238000013528 artificial neural network Methods 0.000 claims description 25
- 238000012360 testing method Methods 0.000 claims description 21
- 230000007704 transition Effects 0.000 claims description 18
- 238000010606 normalization Methods 0.000 claims description 12
- 238000012937 correction Methods 0.000 claims description 10
- 238000011156 evaluation Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 7
- 239000002356 single layer Substances 0.000 claims description 7
- 239000004973 liquid crystal related substance Substances 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000012417 linear regression Methods 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 238000013527 convolutional neural network Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 5
- 230000036962 time dependent Effects 0.000 description 5
- 125000004122 cyclic group Chemical group 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000036541 health Effects 0.000 description 4
- 230000000306 recurrent effect Effects 0.000 description 4
- 230000008034 disappearance Effects 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 3
- 230000002829 reductive effect Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 235000017166 Bambusa arundinacea Nutrition 0.000 description 1
- 235000017491 Bambusa tulda Nutrition 0.000 description 1
- 241001330002 Bambuseae Species 0.000 description 1
- 235000015334 Phyllostachys viridis Nutrition 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 239000011425 bamboo Substances 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- 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
- 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
- G06N3/0455—Auto-encoder networks; Encoder-decoder 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a method, a device, electronic equipment and a medium for predicting the residual service life of an engine, wherein the method comprises the following steps: establishing an initial engine remaining useful life prediction model, wherein the initial engine remaining useful life prediction model follows an encoder-decoder structure and the encoder includes an hourglass feature extractor and a masked self-attention mechanism; and acquiring degradation data of the engine to be tested, and determining the residual service life of the engine to be tested based on a complete training engine residual service life prediction model. The degradation data is scaled to a plurality of scales through the hourglass-shaped feature extractor to perform feature fusion, and masking processing is performed through a self-attention mechanism with masking, so that time correlation features on different scales are obtained, the reliability of the data features is further improved, and the accuracy of the residual service life prediction model of the engine on long-time sequence prediction is further improved.
Description
Technical Field
The invention relates to the technical field of aeroengine fault prediction and health management, in particular to a method and a device for predicting the residual service life of an engine, electronic equipment and a medium.
Background
As manufacturing and industrial systems mature, various mechanical equipment and devices play a very important role for various industries. Aeroengines, as a critical component of an aircraft, can cause significant disasters if suddenly fails during operation. The predictive health management technology (PHM) of devices is a good management scheme for devices. The prediction of the remaining service life of an aeroengine is an important one in PHM technology.
In order to obtain the remaining service life of the apparatus, the commonly used remaining service life prediction methods are roughly classified into three types: a method based on a traditional physical model; a data driven method; the two methods are mixed. Because it is difficult to build accurate dynamic model based on physical model, and hybrid method is difficult to implement, data-driven method is often adopted, and deep learning method is also the main stream in data-driven method. However, in the existing life prediction method based on deep learning, most of time-dependent information in a time sequence is extracted by using a Convolutional Neural Network (CNN) or a cyclic neural network (RNN), but these networks have poor long-term time-dependent extraction effects on feature data, and the recursive structure of the RNN has problems such as gradient disappearance.
Therefore, in the prior art, in the process of predicting the residual service life of the engine, the prediction accuracy of the long-time sequence is low.
Disclosure of Invention
In view of the foregoing, it is necessary to provide a method, an apparatus, an electronic device and a medium for predicting the remaining service life of an engine, so as to solve the problem of low prediction accuracy in the process of predicting the remaining service life of the engine in the prior art.
In order to solve the above problems, the present invention provides a method for predicting remaining service life of an engine, comprising:
acquiring a degradation data sample of a sample engine;
establishing an initial engine remaining useful life prediction model, wherein the initial engine remaining useful life prediction model follows an encoder-decoder structure and the encoder includes an hourglass feature extractor and a masked self-attention mechanism;
the degradation data sample is input into an initial engine residual service life prediction model, data characteristic extraction is carried out on the degradation data sample based on an hourglass-shaped characteristic extractor, the residual service life sample of the sample engine is taken as output, and the initial engine residual service life prediction model is trained to obtain a complete engine residual service life prediction model;
and acquiring degradation data of the engine to be tested, and determining the residual service life of the engine to be tested based on a complete training engine residual service life prediction model.
Further, obtaining a degradation data sample of the sample engine includes:
acquiring initial degradation data of a sample engine;
deleting constant degradation data in the initial degradation data to obtain transitional degradation data;
and carrying out normalization processing and correction processing on the transition degradation data to obtain a degradation data sample.
Further, the normalizing and correcting process is performed on the transitional degradation data, and the method further comprises the following steps:
setting a linear regression model to train the transition degradation data to obtain regression coefficient estimation of the transition degradation data;
acquiring a sequence average value of transition degradation data;
and splicing the transition degradation data with the regression coefficient estimation and the sequence average value respectively.
Further, the initial engine residual service life prediction model comprises an input module, an encoding module, a decoding module and an output module;
the input module comprises a linear layer and a position information coding unit;
the coding module comprises an hourglass-shaped feature extractor, a three-layer self-attention mechanism with a shade and a feedforward neural network;
the decoding module comprises a single-layer multi-head attention mechanism;
the output module includes a flattening layer unit, two linear layer units, and an activation function.
Further, the method for inputting the degradation data sample into the initial engine residual service life prediction model, extracting the data characteristic of the degradation data sample based on the hourglass-shaped characteristic extractor, and taking the residual service life sample of the sample engine as output comprises the following steps:
inputting the degradation data sample to an input module, and processing the degradation data sample through a linear layer and a position information coding unit to obtain a linear degradation data sample;
scaling the linear degradation data sample through an hourglass-shaped feature extractor, and fusing features with different scales to obtain a multi-scale fusion feature map;
inputting the multi-scale fusion feature map to a three-layer self-attention mechanism with a shade, obtaining time correlation features of different scales, and connecting the time correlation features by a feedforward neural network to obtain a coding feature map;
inputting the coding feature map to a decoder, and inquiring and focusing based on a single-layer multi-head attention mechanism to obtain a focusing result vector;
and inputting the attention result vector to an output module, and sequentially passing through the flattening layer unit, the two linear layer units and the activation function to obtain a residual service life predicted value sample.
Further, the hourglass-shaped feature extractor comprises two downsampling convolution layers, two upsampling convolution layers, and a 1×1 one-dimensional convolution layer; extracting data characteristics of the linear degradation data samples through an hourglass-shaped characteristic extractor, masking through a three-layer self-attention mechanism with masks, and connecting data through a feedforward neural network to obtain residual service life coding samples, wherein the method comprises the following steps of:
sequentially passing the linear degradation data sample through two downsampling convolution layers and two upsampling convolution layers to respectively obtain upsampling characteristic data and downsampling characteristic data;
according to the one-dimensional convolution layer, corresponding addition fusion is carried out on the data with equal time steps in the up-sampling characteristic data and the down-sampling characteristic data, and normalization processing is carried out, so that fusion characteristic data is obtained;
and inputting the fusion characteristic data into a three-layer self-attention mechanism mask with a mask, and performing data connection by a feedforward neural network to obtain a residual service life coding sample.
Further, the three-layer masked self-attention mechanism includes a three-layer pyramid self-attention layer and a masked self-attention layer; inputting the multi-scale fusion feature map to a three-layer self-attention mechanism with a shade, obtaining time correlation features of different scales, and connecting the time correlation features by a feedforward neural network to obtain a coding feature map, wherein the method comprises the following steps of:
inputting the multi-scale fusion feature map to a three-layer pyramid self-attention layer, carrying out mask processing on the multi-scale fusion feature map by the self-attention layer with a mask, selecting to obtain a child node, an A node and a father node, and setting the rest nodes to be minus infinity;
the principle formula of masking the multi-scale fusion feature map by the self-attention layer with the mask is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
respectively representing all A nodes, child nodes and father nodes corresponding to the first data point of the s layer,/->
Represents the s-th and j-th nodes, A, C, P represents the number of child nodes and parent nodes of the A node, and s represents the number of layers.
Further, obtaining a trained complete engine residual service life prediction model, further comprises:
establishing an evaluation index, and evaluating the test result of the residual service life prediction model of the engine;
according to the evaluation result, adjusting relevant parameters of the residual service life prediction model of the engine to obtain a completely trained residual service life prediction model of the engine;
wherein, the evaluation index includes:
wherein n represents the total number of aero-engines, RUL pred，i Representing a residual life prediction value, RUL, of an ith aircraft engine new，i Indicating the remaining service life of the first aeroengine after the service life correction.
In order to solve the above-mentioned problems, the present invention also provides an engine remaining life prediction apparatus comprising:
the sample acquisition module is used for acquiring a degradation data sample of the sample engine;
a model building module for building an initial engine remaining useful life prediction model, wherein the initial engine remaining useful life prediction model follows an encoder-decoder structure, and the encoder comprises an hourglass feature extractor and a self-attention mechanism with a mask;
the model training module is used for inputting the degradation data sample into an initial engine residual service life prediction model, extracting data characteristics of the degradation data sample based on the hourglass-shaped characteristic extractor, taking the residual service life sample of the sample engine as output, and training the initial engine residual service life prediction model to obtain a completely trained engine residual service life prediction model;
the remaining service life determining module is used for acquiring degradation data of the engine to be tested and determining the remaining service life of the engine to be tested based on a complete training engine remaining service life prediction model.
In order to solve the above-mentioned problem, the present invention further provides an electronic device, including a processor and a memory, where the memory stores a computer program, and when the computer program is executed by the processor, the method for predicting remaining service life of an engine is implemented as described above.
The beneficial effects of adopting above-mentioned technical scheme are: the invention provides a method, a device, electronic equipment and a medium for predicting the residual service life of an engine, wherein the method comprises the following steps: establishing an initial engine remaining useful life prediction model, wherein the initial engine remaining useful life prediction model follows an encoder-decoder structure and the encoder includes an hourglass feature extractor and a masked self-attention mechanism; and acquiring degradation data of the engine to be tested, and determining the residual service life of the engine to be tested based on a complete training engine residual service life prediction model. The degradation data is scaled to a plurality of scales through the hourglass-shaped feature extractor to perform feature fusion, and masking processing is performed through a self-attention mechanism with masking, so that time correlation features on different scales are obtained, the reliability of the data features is further improved, and the accuracy of the residual service life prediction model of the engine on long-time sequence prediction is further improved.
Drawings
FIG. 1 is a flow chart of an embodiment of a method for predicting remaining service life of an engine according to the present invention;
FIG. 2 is a flow chart of an embodiment of obtaining a degradation data sample of a sample engine according to the present invention;
FIG. 3 is a flow chart of an embodiment of a predictive model for training a residual life of an initial engine according to the present invention;
FIG. 4 is a flow chart of an embodiment of obtaining a remaining life code sample according to the present invention;
FIG. 5 is a schematic diagram of a device for predicting remaining service life of an engine according to the present invention;
fig. 6 is a block diagram of an embodiment of an electronic device according to the present invention.
Detailed Description
Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form a part hereof, and together with the description serve to explain the principles of the invention, and are not intended to limit the scope of the invention.
Before describing the embodiments, a convolutional neural network and a cyclic neural network are described:
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, the three key operations are local receptive field, weight sharing and a pooling layer, so that the number of parameters of the neural network can be effectively reduced, and the overfitting problem of a model is relieved.
The recurrent neural network (Recurrent Neural Network, RNN) is a type of recurrent neural network (recursive neural network) that takes sequence data as input, performs recursion (recovery) in the evolution direction of the sequence, and all nodes (circulation units) are chained.
In the gradient calculation method in the cyclic neural network, the problems of gradient disappearance, error accumulation and the like easily occur when the time step is long. With this limitation, it is difficult for recurrent neural networks to capture time-dependent features between long-time series elements.
The fault prediction and health management (Prognostics Health Management, PHM) is the upgrade and development based on state maintenance (Condition based maintenance, CBM) proposed for meeting the requirement of autonomous guarantee and autonomous diagnosis of mechanical equipment, and is firstly applied to the field of aeroengines, and along with the development of industrial systems and PHM technology, bamboo slips are applied to the fields of medical treatment, military equipment and the like. The residual life prediction of the aeroengine belongs to the PHM technology, and the technology is applied to predicting equipment aging faults and the like possibly occurring in the aeroengine so as to more efficiently maintain the engine, and has important significance for improving the reliability, safety and maintenance efficiency of the aeroengine.
In general, methods for predicting remaining life of an aircraft engine can be broadly divided into three types: traditional model-based methods, data-driven methods, and hybrid methods of both. Among them, the traditional model-based method requires accurate dynamic modeling of mechanical equipment to describe the degradation trend of the component, and complicated and accurate modeling of the modern equipment structure is difficult to realize. Thus, data driven methods are evolving faster. Meanwhile, as the industrial data volume is larger and larger, the deep learning method is adopted in the data driving method to automatically extract the characteristics and develop rapidly with higher accuracy.
In life prediction methods based on deep learning, most of the life prediction methods use Convolutional Neural Networks (CNNs) or cyclic neural networks (RNNs) to extract time-dependent information in time series, but these networks have poor long-term time-dependent extraction effects on feature data, and the recursive structure of RNNs has the problem of gradient disappearance and the like. Therefore, in the prior art, in the process of predicting the remaining service life of the engine, there is a problem of low prediction accuracy.
In order to solve the problem of low prediction accuracy in the process of predicting the residual service life of an engine in the prior art, the invention provides a method, a device, electronic equipment and a storage medium for predicting the residual service life of the engine, and the method, the device, the electronic equipment and the storage medium are respectively described in detail below.
Referring to fig. 1, fig. 1 is a schematic flow chart of an embodiment of a method for predicting remaining service life of an engine according to the present invention, including:
step S101: acquiring a degradation data sample of a sample engine;
step S102: establishing an initial engine remaining useful life prediction model, wherein the initial engine remaining useful life prediction model follows an encoder-decoder structure and the encoder includes an hourglass feature extractor and a masked self-attention mechanism;
step S103: the degradation data sample is input into an initial engine residual service life prediction model, data characteristic extraction is carried out on the degradation data sample based on an hourglass-shaped characteristic extractor, the residual service life sample of the sample engine is taken as output, and the initial engine residual service life prediction model is trained to obtain a complete engine residual service life prediction model;
step S104: and acquiring degradation data of the engine to be tested, and determining the residual service life of the engine to be tested based on a complete training engine residual service life prediction model.
In this embodiment, first, a degradation data sample of a sample engine is obtained; secondly, establishing an initial engine residual service life prediction model, wherein the initial engine residual service life prediction model follows an encoder-decoder structure, and the encoder comprises an hourglass-shaped feature extractor and a self-attention mechanism with a mask; then, inputting the degradation data sample into an initial engine residual service life prediction model, extracting data features of the degradation data sample based on an hourglass feature extractor, and training the initial engine residual service life prediction model by taking the residual service life sample of the sample engine as output to obtain a completely trained engine residual service life prediction model; and finally, acquiring degradation data of the engine to be tested, and determining the residual service life of the engine to be tested based on a complete training engine residual service life prediction model.
In this embodiment, an hourglass-shaped feature extractor is set in an encoding module of an engine residual service life prediction model, degradation data is scaled to multiple scales based on the hourglass-shaped feature extractor to perform feature fusion, and masking processing is performed by a self-attention mechanism with masking, so that time correlation features on different scales are obtained, reliability of the data features is further improved, and accuracy of the engine residual service life prediction model on long-time sequence prediction is further improved.
As a preferred embodiment, in step S101, in order to obtain a degradation data sample of a sample engine, as shown in fig. 2, fig. 2 is a flow chart of an embodiment of obtaining a degradation data sample of a sample engine according to the present invention, which includes:
step S111: acquiring initial degradation data of a sample engine;
step S112: deleting constant degradation data in the initial degradation data to obtain transitional degradation data;
step S113: and carrying out normalization processing and correction processing on the transition degradation data to obtain a degradation data sample.
In this embodiment, first, initial degradation data of a sample engine is acquired; then deleting constant degradation data in the initial degradation data to obtain transition degradation data; and finally, carrying out normalization processing and correction processing on the transition degradation data to obtain a degradation data sample.
In the embodiment, constant degradation data which is kept unchanged in the operation period is removed, so that the influence of invalid data on the prediction performance is avoided; by carrying out normalization processing and correction processing on the data, adverse effects caused by singular sample data can be eliminated, the calculation difficulty can be reduced, the data consistency is improved, and the accuracy of a finally obtained prediction result is improved.
In a specific embodiment, in step S112, all initial degradation data of the sample engine are listed, and sensor data that has not changed are removed to obtain transitional degradation data; that is, sensor data which is unchanged in the initial degradation data is removed, and transitional degradation data is obtained.
In a specific embodiment, in step S113, normalization processing and correction processing are further required to be performed on the transitional degradation data, so as to obtain a degradation data sample. In this embodiment, a Min-Max normalization method is adopted to scale the transitional degradation data to the range of [0,1], and the specific formula is as follows:
wherein x is norm The normalized data is represented, min (x) represents the minimum value in time series x, and max (x) represents the maximum value in time series x.
Further, the remaining service life of the engine needs to be corrected, and the maximum service life RUL is set first max The degradation data samples are then determined as follows:
wherein RUL new Representing degraded data samples, RUL origiin And the residual service life value obtained by the original test of the aero-engine is shown.
In one embodiment, the maximum lifetime is set to 125.
Further, to facilitate data processing, the degraded data samples need to be processed at intervals. In a specific embodiment, firstly dividing a degraded data sample, and determining a training set and a testing set; then, the training set and the testing set are intercepted by a sliding time window, wherein the sliding window is 50 in size, the sliding interval is one time step, and the training set and the testing set with the new time step length of 50 are finally obtained through processing.
In other embodiments, the sliding window size may also be adaptively adjusted according to actual needs.
As a preferred embodiment, the transitional degradation data is also subjected to a splicing process before being subjected to a normalization process and a correction process. Firstly, setting a linear regression model to train the transition degradation data to obtain regression coefficient estimation of the transition degradation data; then, obtaining a sequence average value of the transition degradation data; and finally, splicing the transition degradation data with the regression coefficient estimation and the sequence average value respectively.
And the data characteristics of the transitional degradation data are conveniently extracted by splicing the transitional degradation data.
As a preferred embodiment, in step S102, the initial engine remaining life prediction model includes an input module, an encoding module, a decoding module, and an output module; the input module comprises a linear layer and a position information coding unit; the coding module comprises an hourglass-shaped feature extractor, a three-layer self-attention mechanism with a shade and a feedforward neural network; the decoding module comprises a single-layer multi-head attention mechanism; the output module includes a flattening layer unit, two linear layer units, and an activation function.
The method is an improvement based on the existing converter model, and the converter model with the hourglass-shaped feature extractor is constructed to realize the construction of an initial engine residual service life prediction model.
As a preferred embodiment, in step S103, in order to train the initial engine remaining life prediction model, as shown in fig. 3, fig. 3 is a flowchart of an embodiment of training the initial engine remaining life prediction model according to the present invention, including:
step S131: inputting the degradation data sample to an input module, and processing the degradation data sample through a linear layer and a position information coding unit to obtain a linear degradation data sample;
step S132: scaling the linear degradation data sample through an hourglass-shaped feature extractor, and fusing features with different scales to obtain a multi-scale fusion feature map;
step S133: inputting the multi-scale fusion feature map to a three-layer self-attention mechanism with a shade, obtaining time correlation features of different scales, and connecting the time correlation features by a feedforward neural network to obtain a coding feature map;
step S134: inputting the coding feature map to a decoder, and inquiring and focusing based on a single-layer multi-head attention mechanism to obtain a focusing result vector;
step S135: and inputting the attention result vector to an output module, and sequentially passing through the flattening layer unit, the two linear layer units and the activation function to obtain a residual service life predicted value sample.
In this embodiment, first, a degraded data sample is input to an input module, and is processed by a linear layer and a position information encoding unit, so as to obtain a linear degraded data sample; secondly, scaling the linear degradation data sample through an hourglass-shaped feature extractor, and fusing features with different scales to obtain a multi-scale fusion feature map; inputting the multi-scale fusion feature map to a three-layer self-attention mechanism with a shade, obtaining time correlation features of different scales, and connecting the time correlation features by a feedforward neural network to obtain a coding feature map; then, inputting the coding feature map to a decoder, and inquiring and focusing based on a single-layer multi-head attention mechanism to obtain a focusing result vector; and finally, inputting the attention result vector to an output module, and sequentially passing through the flattening layer unit, the two linear layer units and the activation function to obtain a residual service life predicted value sample.
In the embodiment, the data characteristics in the linear degradation data sample are extracted through the hourglass-shaped characteristic extractor, and characteristic fusion is carried out, so that the accuracy and reliability of the data characteristics can be improved; the self-attention mechanism with the three layers of masks is used for masking, so that the training effect of the initial engine residual service life prediction model can be monitored in a targeted manner, and the corresponding parameters of the initial engine residual service life prediction model are adaptively improved, thereby improving the reliability of the initial engine residual service life prediction model.
In a preferred embodiment, in step S132, the hourglass-shaped feature extractor includes two downsampled convolution layers, two upsampled convolution layers, and a 1×1 one-dimensional convolution layer, and in order to obtain the remaining life code samples, as shown in fig. 4, fig. 4 is a schematic flow chart of an embodiment of obtaining the remaining life code samples according to the present invention, including:
step S1321: sequentially passing the linear degradation data sample through two downsampling convolution layers and two upsampling convolution layers to respectively obtain upsampling characteristic data and downsampling characteristic data;
step S1322: according to the one-dimensional convolution layer, corresponding addition fusion is carried out on the data with equal time steps in the up-sampling characteristic data and the down-sampling characteristic data, and normalization processing is carried out, so that fusion characteristic data is obtained;
step S1323: and inputting the fusion characteristic data into a three-layer self-attention mechanism mask with a mask, and performing data connection by a feedforward neural network to obtain a residual service life coding sample.
In one embodiment, in step S1323, the three-layer masked self-attention mechanism includes a three-layer pyramid self-attention layer and a masked self-attention layer; inputting the multi-scale fusion feature map to a three-layer pyramid self-attention layer, carrying out mask processing on the multi-scale fusion feature map by the self-attention layer with a mask, selecting to obtain a child node, an A node and a father node, and setting the rest nodes to be minus infinity;
the principle formula of masking the multi-scale fusion feature map by the self-attention layer with the mask is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
respectively representing all A nodes, child nodes and father nodes corresponding to the first data point of the s layer,/->
Represents the s-th and j-th nodes, A, C, P represents the number of child nodes and parent nodes of the A node, and s represents the number of layers.
In this embodiment, by selecting adjacent nodes under different scales and setting the other nodes to minus infinity, and setting the adjacent nodes to zero after the softmax function, the computational complexity can be reduced, and meanwhile, the attention between adjacent nodes of different scales can be improved.
In other embodiments, the parameters may be adaptively adjusted according to actual needs.
Further, in order to evaluate the training result of the engine residual service life prediction model, an evaluation index is also established to evaluate the prediction result, and then, according to the evaluation result, the relevant parameters of the engine residual service life prediction model are adjusted to obtain the engine residual service life prediction model with complete training.
Wherein, the evaluation index includes:
wherein n represents the total number of aero-engines, RUL pred，i Representing a residual life prediction value, RUL, of an ith aircraft engine new，i Indicating the i-th aircraft engine passAnd (5) remaining service life after the service life correction.
Finally, in order to clearly explain the prediction effect by the engine remaining life prediction model in the present application in detail, the present application is described by the following specific examples.
The NASA-provided C-MAPSS dataset for an aviation turbofan engine is employed in this embodiment and contains four sub-datasets (FD 001 to FD 004). Each sub-dataset has engine history degradation data for different operating conditions and failure modes. To show the effect, the invention uses the FD004 data set with the highest prediction difficulty. The FD004 dataset included a training set of 249 engines complete degradation to failure data and a test set of 248 engines starting degradation to pre-failure random time intercept data acquisition.
First, the changes listed for the data in the dataset contained an engine number, 3 operating condition settings, 21 sensor data. Wherein the sensor data of No. 1,5, 10, 16, 18, 19 are hardly changed, and the corresponding column data is removed; and then carrying out integration classification on the 3 running condition settings of each engine to obtain 6 running conditions, and carrying out Min-max normalization on the remaining 15 sensor data of the 249 engines according to the setting conditions in 6. The data lifetime is then modified to 125. The data for each engine of the training set is then truncated using a sliding window method to obtain training data of 50 x 15 in the final two dimensions. For the test set, each engine intercepts the first 50 data as the test set, and for the engine data with the step length smaller than the sliding window, a linear interpolation method is adopted to insert the data step length to 50 so as to obtain 248 multiplied by 50 multiplied by 15.
And secondly, placing the obtained sensor data of the training set and the testing set into a linear regression model for preliminary feature extraction, and taking the average value of the corresponding extracted linear regression coefficient and the corresponding sensor data of each engine in the data set as new feature data and splicing the new feature data with the original data, wherein the time step dimension is changed from 50 to 52.
Then, an initial engine residual service life prediction model is constructed, training data and test data are input into the model, a training prediction result is obtained through the training data, the prediction result is compared with the actual service residual service life to calculate training loss, parameters of the model are optimized and updated through an ADAM optimizer through the training loss, then a test result is obtained through a test set, and the test loss and the test score are calculated through the test result and the actual value to evaluate the prediction performance of the current model.
Finally, setting the initial optimal loss as a value far greater than the test loss, comparing the test loss with the optimal loss every iteration, and updating the optimal loss into the current test loss and storing the current model if the test loss is smaller than the optimal loss; if the test loss is greater than the optimal loss, the learning rate attenuation strategy is entered, namely whether the current loss is within the range of the optimal loss plus a threshold value is judged, if the current loss is normal jump, the next iteration is entered, if the current loss is out of the range, the learning rate of the optimizer is updated to be smaller, meanwhile, the early stop strategy is started, namely the stop count is increased by one, and when the stop count reaches the set threshold value 5, the current stored model is derived as the optimal model.
Further, the effect of the method can be evaluated by comparing the prediction accuracy of the method with other methods, and the RMSE and Score evaluation results of the method and other methods are shown in the following table:
in sum, by arranging the hourglass-shaped feature extractor in the coding module of the residual service life prediction model of the engine, the degradation data of the engine are processed based on the hourglass-shaped feature extractor, and the data features are fused, so that the data features in the degradation data can be effectively extracted, the reliability of the data features is further improved, and the accuracy of the residual service life prediction model of the engine to long-time sequence prediction is further improved.
In order to solve the above-mentioned problems, the present invention further provides an apparatus for predicting remaining service life of an engine, as shown in fig. 5, fig. 5 is a schematic structural diagram of the apparatus for predicting remaining service life of an engine, and the apparatus 500 for predicting remaining service life of an engine includes:
a sample acquisition module 501 for acquiring a degradation data sample of a sample engine;
a model building module 502 for building an initial engine remaining useful life prediction model, wherein the initial engine remaining useful life prediction model follows an encoder-decoder structure and the encoder includes an hourglass feature extractor and a masked self-attention mechanism;
the model training module 503 is configured to input the degradation data sample to the initial engine residual service life prediction model, perform data feature extraction on the degradation data sample based on the hourglass feature extractor, and train the initial engine residual service life prediction model with the residual service life sample of the sample engine as output, to obtain a completely trained engine residual service life prediction model;
the remaining service life determining module 504 is configured to obtain degradation data of an engine to be tested, and determine a remaining service life of the engine to be tested based on the trained complete engine remaining service life prediction model.
The invention also correspondingly provides an electronic device, as shown in fig. 6, and fig. 6 is a structural block diagram of an embodiment of the electronic device provided by the invention. The electronic device 600 may be a computing device such as a mobile terminal, desktop computer, notebook, palm top computer, server, etc. The electronic device 600 comprises a processor 601 and a memory 602, wherein the memory 602 has stored thereon a remaining life prediction program 603 of the engine.
The memory 602 may be an internal storage unit of a computer device in some embodiments, such as a hard disk or memory of a computer device. The memory 602 may also be an external storage device of the computer device in other embodiments, such as a plug-in hard disk provided on the computer device, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), etc. Further, the memory 602 may also include both internal storage units and external storage devices of the computer device. The memory 602 is used for storing application software installed on the computer device and various types of data, such as program codes for installing the computer device. The memory 602 may also be used to temporarily store data that has been output or is to be output. In one embodiment, the engine remaining life prediction program 603 may be executed by the processor 601 to implement the engine remaining life prediction method of the embodiments of the present invention.
The processor 601 may in some embodiments be a central processing unit (Central Processing Unit, CPU), microprocessor or other data processing chip for executing program code or processing data stored in the memory 602, such as executing an engine remaining life prediction program or the like.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the various embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile 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), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present invention are intended to be included in the scope of the present invention.
Claims (10)
1. A method for predicting remaining service life of an engine, comprising:
acquiring a degradation data sample of a sample engine;
establishing an initial engine remaining useful life prediction model, wherein the initial engine remaining useful life prediction model follows an encoder-decoder structure, and the encoder comprises an hourglass feature extractor and a masked self-attention mechanism;
inputting the degradation data sample into the initial engine residual service life prediction model, extracting data features of the degradation data sample based on the hourglass feature extractor, and training the initial engine residual service life prediction model by taking the residual service life sample of the sample engine as output to obtain a completely trained engine residual service life prediction model;
and acquiring degradation data of the engine to be tested, and determining the residual service life of the engine to be tested based on the trained complete engine residual service life prediction model.
2. The method of claim 1, wherein obtaining a sample of degradation data for a sample engine comprises:
acquiring initial degradation data of a sample engine;
deleting constant degradation data in the initial degradation data to obtain transition degradation data;
and carrying out normalization processing and correction processing on the transition degradation data to obtain a degradation data sample.
3. The method for predicting remaining service life of an engine according to claim 1, wherein the normalizing and correcting process is performed on the transitional degradation data, further comprising:
setting a linear regression model to train the transition degradation data to obtain regression coefficient estimation of the transition degradation data;
acquiring a sequence average value of the transition degradation data;
and respectively splicing the transition degradation data with the regression coefficient estimation and the sequence average value.
4. The method of claim 1, wherein the initial engine remaining life prediction model comprises an input module, an encoding module, a decoding module, and an output module;
the input module comprises a linear layer and a position information coding unit;
the coding module comprises an hourglass-shaped feature extractor, a three-layer self-attention mechanism with a shade and a feedforward neural network;
the decoding module comprises a single-layer multi-head attention mechanism;
the output module includes a flattening layer unit, two linear layer units, and an activation function.
5. The engine remaining life prediction method according to claim 4, wherein inputting the degradation data sample into the initial engine remaining life prediction model, performing data feature extraction on the degradation data sample based on the hourglass feature extractor, and taking a remaining life sample of the sample engine as an output, comprises:
inputting the degradation data sample to the input module, and processing the degradation data sample through the linear layer and the position information coding unit to obtain a linear degradation data sample;
scaling the linear degradation data sample through the hourglass-shaped feature extractor, and fusing features with different scales to obtain a multi-scale fusion feature map;
inputting the multi-scale fusion feature map to a self-attention mechanism of the three layers of masking, obtaining time correlation features of different scales, and connecting the time correlation features by the feedforward neural network to obtain a coding feature map;
inputting the coding feature map to the decoder, and inquiring and focusing based on the single-layer multi-head attention mechanism to obtain a focusing result vector;
and inputting the attention result vector to the output module, and sequentially passing through the flattening layer unit, the two linear layer units and the activation function to obtain the residual service life predicted value sample.
6. The method of claim 5, wherein the hourglass feature extractor comprises two downsampled convolution layers, two upsampled convolution layers, and a 1 x 1 one-dimensional convolution layer; extracting data features of the linear degradation data samples through the hourglass-shaped feature extractor, masking through a self-attention mechanism of the three layers of masking, and performing data connection through the feedforward neural network to obtain residual service life coding samples, wherein the method comprises the following steps of:
sequentially passing the linear degradation data sample through the two downsampling convolution layers and the two upsampling convolution layers to respectively obtain upsampling characteristic data and downsampling characteristic data;
according to the one-dimensional convolution layer, corresponding addition fusion is carried out on the data with equal time step in the up-sampling characteristic data and the down-sampling characteristic data, and normalization processing is carried out, so that fusion characteristic data is obtained;
and inputting the fusion characteristic data to the three-layer self-attention mechanism shade with the shade, and carrying out data connection by the feedforward neural network to obtain a residual service life coding sample.
7. The method of claim 6, wherein the three-layer masked self-attention mechanism comprises a three-layer pyramid self-attention layer and a masked self-attention layer; inputting the multi-scale fusion feature map to a self-attention mechanism of the three layers of masking, obtaining time correlation features of different scales, and connecting the time correlation features by the feedforward neural network to obtain a coding feature map, wherein the method comprises the following steps of:
inputting the multi-scale fusion feature map to the three-layer pyramid self-attention layer, firstly carrying out mask processing on the multi-scale fusion feature map by the self-attention layer with the mask, selecting to obtain a child node, an A node and a father node, and setting the rest nodes to be minus infinity;
the principle formula of masking the multi-scale fusion feature map by the self-attention layer with the mask is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
P l (s) respectively representing all A nodes, child nodes and father nodes corresponding to the first data point of the s layer,/->
Represents the s-th and j-th nodes, A, C, P represents the number of child nodes and parent nodes of the A node, and s represents the number of layers.
8. The method of claim 1, wherein obtaining a trained complete engine remaining life prediction model, further comprises:
establishing an evaluation index, and evaluating a test result of the residual service life prediction model of the engine;
according to the evaluation result, adjusting relevant parameters of the residual service life prediction model of the engine to obtain a completely trained residual service life prediction model of the engine;
wherein the evaluation index includes:
wherein n represents the total number of aero-engines, RUL pred，i Representing a residual life prediction value, RUL, of an ith aircraft engine new，i And the residual service life of the ith aeroengine after the service life correction is represented.
9. An engine remaining life prediction apparatus, comprising:
the sample acquisition module is used for acquiring a degradation data sample of the sample engine;
a model building module for building an initial engine remaining useful life prediction model, wherein the initial engine remaining useful life prediction model follows an encoder-decoder structure, and the encoder comprises an hourglass feature extractor and a masked self-attention mechanism;
the model training module is used for inputting the degradation data sample into the initial engine residual service life prediction model, extracting data characteristics of the degradation data sample based on the hourglass-shaped characteristic extractor, and training the initial engine residual service life prediction model by taking the residual service life sample of the sample engine as output to obtain a completely trained engine residual service life prediction model;
the remaining service life determining module is used for acquiring degradation data of the engine to be tested and determining the remaining service life of the engine to be tested based on the trained complete engine remaining service life prediction model.
10. An electronic device comprising a processor and a memory, the memory having stored thereon a computer program which, when executed by the processor, implements the method of predicting remaining useful life of an engine as claimed in any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310270179.1A CN116307183A (en) | 2023-03-15 | 2023-03-15 | Engine residual service life prediction method and device, electronic equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310270179.1A CN116307183A (en) | 2023-03-15 | 2023-03-15 | Engine residual service life prediction method and device, electronic equipment and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116307183A true CN116307183A (en) | 2023-06-23 |
Family
ID=86837611
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310270179.1A Pending CN116307183A (en) | 2023-03-15 | 2023-03-15 | Engine residual service life prediction method and device, electronic equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116307183A (en) |
-
2023
- 2023-03-15 CN CN202310270179.1A patent/CN116307183A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111210024B (en) | Model training method, device, computer equipment and storage medium | |
| US20220358363A1 (en) | Engine surge fault prediction system and method based on fusion neural network model | |
| US11544676B2 (en) | Equipment repair management and execution | |
| CN108780315B (en) | Method and apparatus for optimizing diagnostics of rotating equipment | |
| US20220187819A1 (en) | Method for event-based failure prediction and remaining useful life estimation | |
| CN113196314B (en) | Adapting a predictive model | |
| CN112257755B (en) | Method and device for analyzing running state of spacecraft | |
| CN111985155B (en) | Circuit health state prediction method and system based on integrated deep neural network | |
| CN112613226B (en) | Feature enhancement method for residual life prediction | |
| CN105593864B (en) | Analytical device degradation for maintenance device | |
| CN115017826B (en) | Method for predicting residual service life of equipment | |
| Zhao et al. | Probabilistic remaining useful life prediction based on deep convolutional neural network | |
| CN113284002A (en) | Power consumption data anomaly detection method and device, computer equipment and storage medium | |
| Akpudo et al. | An automated sensor fusion approach for the rul prediction of electromagnetic pumps | |
| Li et al. | Prognosis for stochastic degrading systems with massive data: A data-model interactive perspective | |
| CN113673774A (en) | Aero-engine remaining life prediction method based on self-encoder and time sequence convolution network | |
| CN115879248B (en) | Full life cycle management method and system suitable for vacuum pump | |
| CN116665798A (en) | Air pollution trend early warning method and related device | |
| CN116307183A (en) | Engine residual service life prediction method and device, electronic equipment and medium | |
| CN116562120A (en) | RVE-based turbine engine system health condition assessment method and RVE-based turbine engine system health condition assessment device | |
| CN112560252B (en) | Method for predicting residual life of aeroengine | |
| CN112598082B (en) | Method and system for predicting generalized error of image identification model based on non-check set | |
| Javanmardi et al. | Conformal Prediction Intervals for Remaining Useful Lifetime Estimation | |
| CN115017819A (en) | Engine remaining service life prediction method and device based on hybrid model | |
| CN116089822B (en) | Equipment RUL prediction method and system based on space-time attention network |
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 |