CN114880767A - Aero-engine remaining service life prediction method based on attention mechanism Dense-GRU network - Google Patents
Aero-engine remaining service life prediction method based on attention mechanism Dense-GRU network Download PDFInfo
- Publication number
- CN114880767A CN114880767A CN202210434877.6A CN202210434877A CN114880767A CN 114880767 A CN114880767 A CN 114880767A CN 202210434877 A CN202210434877 A CN 202210434877A CN 114880767 A CN114880767 A CN 114880767A
- Authority
- CN
- China
- Prior art keywords
- layer
- sample
- model
- output
- service life
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 230000007246 mechanism Effects 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000006731 degradation reaction Methods 0.000 claims description 36
- 230000015556 catabolic process Effects 0.000 claims description 27
- 230000004913 activation Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000008034 disappearance Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 17
- 230000008569 process Effects 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 208000006096 Attention Deficit Disorder with Hyperactivity Diseases 0.000 description 1
- 208000036864 Attention deficit/hyperactivity disease Diseases 0.000 description 1
- 208000015802 attention deficit-hyperactivity disease Diseases 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Abstract
The invention discloses an aeroengine remaining service life prediction method based on an attention mechanism Dense-GRU network, which comprises the following steps: step one, data preprocessing; step two, constructing an attention mechanism layer; step three, constructing and training an integral model; fourthly, predicting the residual service life by using the model; the method provides a density-GRU network based on an attention mechanism, and the attention mechanism is added to the multidimensional sensor data of the engine in the provided model, so that the network model can pay more attention to the more effective sensors for prediction. The network main part is a Dense-GRU network, so that the propagation and reuse of multidimensional time series characteristics are enhanced, the problem of gradient disappearance of a time series deep network is effectively avoided, and the accuracy of residual service life prediction is effectively improved.
Description
Technical Field
The invention belongs to the technical field of prediction of the remaining service life of an aircraft engine, and particularly relates to a method for predicting the remaining service life of the aircraft engine based on a attention mechanism Dense-GRU network.
Background
The residual service life prediction is to estimate the residual service life of the engine by analyzing the trend of the historical performance degradation of the engine, and the accurate residual service prediction can improve the reliability and safety of equipment or a system, prevent fatal faults, reduce maintenance cost and provide support for an airline company to establish a long-term maintenance plan of an engine fleet.
Currently, the remaining life prediction is divided into two directions: based on physical degradation models and based on data driving. The residual life prediction is performed by mostly using a mathematical process based on a physical degradation model, for example, algebraic equations and differential equations are used to describe the degradation process. The prior knowledge dependency of the method on the equipment degradation process is too high, and due to the complexity and randomness of the system degradation process, accurate description and modeling through a mathematical model are difficult. A data-driven-based method starts from monitoring data of a system state, and a model is established to explore potential relations between the monitoring data of the sensor and the system degradation state. With the development of computer and sensor technologies, data-driven-based methods are becoming mainstream in the field of residual service life prediction research, which can effectively reduce research costs and mine complex relationships between degraded data and residual service life.
With the great success of deep learning in the field of computer vision, the research method for predicting the remaining service life gradually changes from the traditional machine learning method to the deep learning method, and the neural network method gradually becomes the mainstream of the data-driven method for predicting the remaining service life. At present, the neural network has the following problems in predicting the residual service life: (1) the monitoring data of the aircraft engine are time sequence data of a plurality of sensors, and the importance degree among the sensors is often ignored when the time sequence of a plurality of sensors is processed, so that the accuracy of prediction is reduced. (2) Deep neural networks mostly adopt a deep structure with multiple layers stacked, the structure can only use the high-level features extracted from the last layer to predict the residual service life, and the features of other layers are all ignored to cause information loss. In order to solve the problems, a method for predicting the remaining service life of the aircraft engine based on a attention-system Dense-GRU network is provided.
Disclosure of Invention
The invention aims to solve the technical problem of providing a method for predicting the remaining service life of an aircraft engine based on a attention mechanism Dense-GRU network aiming at the defects in the prior art. Firstly, the attention mechanism is added into the multidimensional sensor data of the engine processed by the proposed model, so that the weight of the sensor data which has great influence on the prediction of the residual service life is improved, and the network model can pay more attention to the more effective sensor for prediction. And secondly, the network main part is a Dense-GRU network, so that the propagation and reuse of multidimensional time series characteristics are enhanced, and the gradient disappearance problem of a time series deep network is effectively avoided. The method fully considers the influence of the difference of the sensors on the prediction of the residual service life, fully excavates and utilizes the degradation information of the multi-dimensional sensor sequence, and effectively improves the accuracy of the prediction of the residual service life.
In order to solve the technical problems, the invention adopts the technical scheme that: a method for predicting the remaining service life of an aircraft engine based on a attention mechanism Dense-GRU network is characterized by comprising the following steps:
step one, data preprocessing:
step 101: setting N moments in the engine degradation process, and inputting m sensor degradation sequence data X in the complete degradation process o1 ＝[X 1 ,X 2 ,…,X m ]The degradation sequence of the jth sensor is represented as
Eliminating sensor data with small degree of correlation with the performance degradation process or basically unchanged parameters of the whole degradation period, and finally retaining the degradation data X of the k sensors o2 ＝[X 1 ,X 2 ,…,X k ]K is less than m;
step 102: normalizing the input k sensor degradation data according to a formula
Calculating normalized data X of each sensor nor ＝[X 1 ,X 2 ,…,X k ]Wherein the normalized data of the jth sensor is
Step 103: to the normalized k sensor data X nor ＝[X 1 ,X 2 ,…,X k ]And carrying out training sample division. Dividing the sample by adopting a sliding window method, wherein the width of the sample is a sensor dimension k, the length of the sample is n, and the sample is expressed as X ═ X 1 ,X 2 ,…,X k ]Wherein
The residual life value corresponding to each sample is the residual life value Y of the last moment of the sample RUL ；
Step two, constructing an attention mechanism layer:
step 201: according to the formula h 1 ＝ReLU(W 1 ·X+b 1 ) Establishing a first linear layer of attention-control layers, wherein X ═ X 1 ,X 2 ,…,X k ]To input samples, h 1 Is the output of the first linear layer, W 1 And b 1 Representing the weight and the bias of the first linear layer, and adopting a ReLU function as an activation function;
step 202: according to the formula h 2 ＝ReLU(W 2 ·h 1 +b 2 ) Establishing a second linear layer of the attention-suppressing layer, wherein h 1 Is the output of the first linear layer, h 2 Is the output of the second linear layer, W 2 And b 2 Representing the weight and the bias of the second linear layer, and adopting a ReLU function as an activation function;
step 203: according to the formula W x ＝softmax(h 2 ) Establishing a weight generation layer for the attention layer, wherein h 2 Is the output of the second linear layer, W x Representing the generated weight, and adopting a softmax function as a calculation function; according to formula X att ＝W x X weights the initial input samples with the generated weights, and finally generates an output X of the attention mechanism layer att ；
Step three, constructing an integral model and training:
step 301: the whole model comprises L-layer GRU unit layers, and adopts denseThe L-layer GRU units are connected in a set connection mode; according to the formula
Calculating the output of the GRU unit of the d-th layer, wherein the input of the current GRU layer is the output of the previous d-1 layer, d is 1,2, … L, and finally calculating the output conversion of the GRU unit of the last layer
Step 302: according to the formula
Calculating to obtain final predicted value of the remaining service life, wherein
To note the output of the force-based Dense-GRU model, FC represents the fully-connected layer,
representing a final predicted value of remaining life;
step 303: according to the formula
Calculating a training loss of the model, wherein
Representing the predicted output of the model, Y t Representing the real output of the samples, and S represents the number of the samples; the model activation function selects a ReLU function, and the optimizer selects an RMSProp optimizer;
step four, using the model to predict the remaining service life:
step 401: let the sample to be predicted be X pre ＝[X 1 ,X 2 ,…,X k ]The degradation sequence of the jth sensor is represented as
Wherein T tableShowing the length of the sample; normalizing the sample data according to step 102, matching the size of the sample with the model input size according to step 103, and generating an input sample X' pre ＝[X 1 ',X' 2 ,…X' k ]The degradation sequence of the jth sensor is represented as
Where n represents the length of the model input sample;
step 402: sample X 'to be detected' pre ＝[X 1 ',X' 2 ,…X' k ]Inputting the predicted residual service life value into the model trained in the third step, and calculating to obtain the residual service life value of the aero-engine at the T +1 moment
Compared with the prior art, the invention has the following advantages:
1. the model provided by the invention adds an attention mechanism in processing the multi-dimensional sensor data, so that the weight of the sensor data which has great influence on the prediction of the residual service life is improved, and the network model can pay more attention to the sensor which is more effective in prediction.
2. The main body part of the network model provided by the invention is based on GRU units, and the Dense connection Dense structure and the GRU units are combined to form a Dense-GRU network, so that the degradation information of a multidimensional sensor sequence is fully mined and utilized, and the propagation and reuse of characteristics are enhanced.
In conclusion, the attention mechanism-based Dense-GRU network model provided by the invention fully considers the influence of the difference of the sensors on the prediction of the residual service life, effectively excavates and utilizes the degradation information of the multidimensional sensor sequence, enhances the propagation and reuse of the characteristics, effectively avoids the gradient disappearance problem of the time sequence deep layer network, and provides a new scheme for the prediction of the residual service life of the aero-engine based on data driving.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
FIG. 1 is a flow chart of the present invention
FIG. 2 is an attention deficit hyperactivity disorder diagram
FIG. 3 is a view showing the overall structure of the model
FIG. 4 is a box plot of the experimental results MAE
FIG. 5 is a sample set generated for the FD001 dataset
FIG. 6 shows the results of different experiments
Detailed Description
The process of the present invention will be described in further detail with reference to examples.
As shown in fig. 1, the present invention comprises the steps of:
step one, data preprocessing:
to verify the validity of the proposed method, a C-MAPSS dataset published by NASA was used for validation. The FD001 dataset of the C-MAPSS dataset was used for training models and testing.
Step 101: to the normalized k sensor data X nor ＝[X 1 ,X 2 ,…,X k ]And carrying out training sample division. Dividing the sample by adopting a sliding window method, wherein the width of the sample is a sensor dimension k, the length of the sample is n, and the sample is expressed as X ═ X 1 ,X 2 ,…,X k ]Wherein
The residual life value corresponding to each sample is the residual life value Y of the last moment of the sample RUL ；
In the actual use process, the window length selected by the sliding window method is the minimum length of the incomplete degradation data of the engines in the test set, namely the length with the shortest degradation period is selected from the incomplete degradation data of 100 aircraft engines in the test set as the sliding window length of the generated sample. In the FD001 dataset, the sensors are finally screened as 16-dimensional sensors, and the shortest length selected for the test set is 31 flight cycles, so the length of the generated sample is 31, the dimension is 16, and the sample set is shown in fig. 5.
Step two, constructing an attention mechanism layer:
step 201: according to the formula h 1 ＝ReLU(W 1 ·X+b 1 ) Establishing a first linear layer of attention bearing layers, wherein X ═ X 1 ,X 2 ,…,X k ]To input samples, h 1 Is the output of the first linear layer, W 1 And b 1 Representing the weight and the bias of the first linear layer, and adopting a ReLU function as an activation function;
step 202: according to the formula h 2 ＝ReLU(W 2 ·h 1 +b 2 ) Establishing a second linear layer of the attention-suppressing layer, wherein h 1 Is the output of the first linear layer, h 2 Is the output of the second linear layer, W 2 And b 2 Representing the weight and the bias of the second linear layer, and adopting a ReLU function as an activation function;
step 203: according to the formula W x ＝softmax(h 2 ) Establishing a weight generation layer for the attention layer, wherein h 2 Is the output of the second linear layer, W x Representing the generated weight, and adopting a softmax function as a calculation function; according to formula X att ＝W x X weights the initial input samples with the generated weights, and finally generates an output X of the attention mechanism layer att ；
In practical use, as shown in fig. 2, the initial input passes through two linear layers with the activation function of ReLU, a series of weights are generated through softmax, and the finally generated weights are weighted with the initial input to serve as the output of the attention mechanism layer.
Step three, calculating the state distance characterization degradation state of the multi-source sensor:
step 301: the overall model comprises L-layer GRU unit layers, and the L-layer GRU units are connected in a dense connection mode; according to the formula
Calculating the output of the GRU unit of the d-th layer, wherein the input of the current GRU layer is the output of the previous d-1 layer, d is 1,2, … L, and finally calculating the output conversion of the GRU unit of the last layer
Step 302: according to the formula
Calculating to obtain a final predicted value of the remaining service life, wherein
To note the output of the force-measure-GRU model, FC denotes the fully-connected layer,
representing a final predicted value of remaining life;
step 303: according to the formula
Calculating a training loss of the model, wherein
Representing the predicted output of the model, Y t Representing the real output of the samples, and S represents the number of the samples; the model activation function selects a ReLU function, and the optimizer selects an RMSProp optimizer;
in the actual use process, the overall model is as shown in fig. 3, the attention mechanism layer is added to the input end of the network model, the GRU network is used as a basic unit in the network, a Dense connection structure is adopted among different GRU unit blocks to form a Dense-GRU, and finally, the final residual service life value is obtained through a full connection layer.
Step four, using the model to predict the remaining service life:
step 401: let the sample to be predicted be X pre ＝[X 1 ,X 2 ,…,X k ]The degradation sequence of the jth sensor is represented as
Wherein T represents the length of the sample; sample pair according to step 102Data is normalized, and the size of the sample is matched to the model input size in step 103 to generate an input sample X' pre ＝[X 1 ',X' 2 ,…X' k ]The degradation sequence of the jth sensor is represented as
Where n represents the length of the model input sample;
step 402: sample X 'to be detected' pre ＝[X 1 ',X' 2 ,…X' k ]Inputting the predicted residual service life value into the model trained in the third step, and calculating to obtain the residual service life value of the aero-engine at the T +1 moment
In the actual use process, an FD001 data set is used for carrying out experiments, in order to verify the effectiveness of the method, classical methods such as a long-time and short-time memory network (LSTM), a gated cyclic unit network (GRU), a bidirectional long-time and short-time memory network (Bi-LSTM) and a residual error network (Resnet) are selected as comparison experiments, the evaluation criterion selects an average absolute error (MAE), a Root Mean Square Error (RMSE) and a Score index, the Score index is defined for a model for predicting the residual life, and the smaller the Score value is, the higher the accuracy of the model is. Table 2 shows the experimental results of the different methods on the test set in FD001 dataset, with the proposed method MAE 18.4415, RMSE 22.9736, and Score 13.9694, with the best results. Fig. 4 is a box diagram of MAE indexes of different models under an FD001 data set, where the box diagram can indicate the error magnitude and dispersion degree of the remaining life prediction results of different models, the rectangular box and the dashed line segment indicate the variation degree of MAE, and the longer the length of the rectangular box and the dashed line segment is, the more dispersed the prediction results of the model are, that is, the worse the prediction stability of the model is, where the line in the middle of the rectangular box indicates the median of MAE, and the smaller the value of the line is, the smaller the prediction error of the model is. From the results, the effectiveness of the present invention was also demonstrated.
The above embodiments are only examples of the present invention, and are not intended to limit the present invention, and all simple modifications, changes and equivalent structural changes made to the above embodiments according to the technical spirit of the present invention still fall within the protection scope of the technical solution of the present invention.
Claims (1)
1. A method for predicting the remaining service life of an aircraft engine based on a attention mechanism Dense-GRU network is characterized by comprising the following steps:
step one, data preprocessing:
step 101: setting N moments in the engine degradation process, and inputting m sensor degradation sequence data X in the complete degradation process o1 ＝[X 1 ,X 2 ,…,X m ]The degradation sequence of the jth sensor is represented as
Eliminating sensor data with small degree of correlation with the performance degradation process or basically unchanged parameters of the whole degradation period, and finally retaining the degradation data X of the k sensors o2 ＝[X 1 ,X 2 ,…,X k ]K is less than m;
step 102: normalizing the input k sensor degradation data according to a formula
Calculating normalized data X of each sensor nor ＝[X 1 ,X 2 ,…,X k ]Wherein the normalized data of the jth sensor is
Step 103: to the normalized k sensor data X nor ＝[X 1 ,X 2 ,…,X k ]And carrying out training sample division. Dividing the sample by adopting a sliding window method, wherein the width of the sample is a sensor dimension k, the length of the sample is n, and the sample is expressed as X ═ X 1 ,X 2 ,…,X k ]Wherein
The residual life value corresponding to each sample is the residual life value Y of the last moment of the sample RUL ；
Step two, constructing an attention mechanism layer:
step 201: according to the formula h 1 ＝ReLU(W 1 ·X+b 1 ) Establishing a first linear layer of attention-control layers, wherein X ═ X 1 ,X 2 ,…,X k ]To input samples, h 1 Is the output of the first linear layer, W 1 And b 1 Representing the weight and the bias of the first linear layer, and adopting a ReLU function as an activation function;
step 202: according to the formula h 2 ＝ReLU(W 2 ·h 1 +b 2 ) Establishing a second linear layer of the attention-suppressing layer, wherein h 1 Is the output of the first linear layer, h 2 Is the output of the second linear layer, W 2 And b 2 Representing the weight and the bias of the second linear layer, and adopting a ReLU function as an activation function;
step 203: according to the formula W x ＝softmax(h 2 ) Establishing a weight generation layer for the attention layer, wherein h 2 Is the output of the second linear layer, W x Representing the generated weight, and adopting a softmax function as a calculation function; according to formula X att ＝W x X weights the initial input samples with the generated weights, and finally generates an output X of the attention mechanism layer att ；
Step three, constructing an integral model and training:
step 301: the overall model comprises L-layer GRU unit layers, and the L-layer GRU units are connected in a dense connection mode; according to the formula
Calculating the output of the GRU unit of the d-th layer, wherein the input of the current GRU layer is the output of the previous d-1 layer, d is 1,2, … L, and finally calculating the output conversion of the GRU unit of the last layer
Step 302: according to the formula
Calculating to obtain a final predicted value of the remaining service life, wherein
To note the output of the force-based Dense-GRU model, FC represents the fully-connected layer,
representing a final predicted value of remaining life;
step 303: according to the formula
Calculating a training loss of the model, wherein Y t pre Representing the predicted output of the model, Y t Representing the real output of the samples, and S represents the number of the samples; the model activation function selects a ReLU function, and the optimizer selects an RMSProp optimizer;
step four, using the model to predict the remaining service life:
step 401: let the sample to be predicted be X pre ＝[X 1 ,X 2 ,…,X k ]The degradation sequence of the jth sensor is represented as
Wherein T represents the length of the sample; normalizing the sample data according to step 102, matching the size of the sample with the model input size according to step 103, and generating an input sample X' pre ＝[X' 1 ,X' 2 ,…X' k ]The degradation sequence of the jth sensor is represented as
Where n represents the length of the model input sample;
step 402: sample X 'to be detected' pre ＝[X' 1 ,X' 2 ,…X' k ]Inputting the predicted residual service life value into the model trained in the third step, and calculating to obtain the residual service life value of the aero-engine at the T +1 moment
。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210434877.6A CN114880767B (en) | 2022-04-24 | 2022-04-24 | Aero-engine residual service life prediction method based on attention mechanism Dense-GRU network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210434877.6A CN114880767B (en) | 2022-04-24 | 2022-04-24 | Aero-engine residual service life prediction method based on attention mechanism Dense-GRU network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114880767A true CN114880767A (en) | 2022-08-09 |
| CN114880767B CN114880767B (en) | 2024-03-08 |
Family
ID=82670736
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210434877.6A Active CN114880767B (en) | 2022-04-24 | 2022-04-24 | Aero-engine residual service life prediction method based on attention mechanism Dense-GRU network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114880767B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116050665A (en) * | 2023-03-14 | 2023-05-02 | 淄博热力有限公司 | Heat supply equipment fault prediction method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110807257A (en) * | 2019-11-04 | 2020-02-18 | 中国人民解放军国防科技大学 | Method for predicting residual life of aircraft engine |
| CN113051689A (en) * | 2021-04-25 | 2021-06-29 | 石家庄铁道大学 | Bearing residual service life prediction method based on convolution gating circulation network |
| US20210406603A1 (en) * | 2020-06-26 | 2021-12-30 | Tata Consultancy Services Limited | Neural networks for handling variable-dimensional time series data |
| CN114186475A (en) * | 2021-10-28 | 2022-03-15 | 南京工业大学 | Pivoting support service life prediction method based on Attention-MGRU |
-
2022
- 2022-04-24 CN CN202210434877.6A patent/CN114880767B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110807257A (en) * | 2019-11-04 | 2020-02-18 | 中国人民解放军国防科技大学 | Method for predicting residual life of aircraft engine |
| US20210406603A1 (en) * | 2020-06-26 | 2021-12-30 | Tata Consultancy Services Limited | Neural networks for handling variable-dimensional time series data |
| CN113051689A (en) * | 2021-04-25 | 2021-06-29 | 石家庄铁道大学 | Bearing residual service life prediction method based on convolution gating circulation network |
| CN114186475A (en) * | 2021-10-28 | 2022-03-15 | 南京工业大学 | Pivoting support service life prediction method based on Attention-MGRU |
Non-Patent Citations (2)
| Title |
|---|
| 王太勇;王廷虎;王鹏;乔卉卉;徐明达;: "基于注意力机制BiLSTM的设备智能故障诊断方法", 天津大学学报(自然科学与工程技术版), no. 06, 27 April 2020 (2020-04-27) * |
| 车畅畅;王华伟;倪晓梅;付强;: "基于改进GRU的航空发动机剩余寿命预测", 航空计算技术, no. 01, 25 January 2020 (2020-01-25) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116050665A (en) * | 2023-03-14 | 2023-05-02 | 淄博热力有限公司 | Heat supply equipment fault prediction method |
| CN116050665B (en) * | 2023-03-14 | 2024-04-02 | 淄博热力有限公司 | Heat supply equipment fault prediction method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114880767B (en) | 2024-03-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112131760B (en) | CBAM model-based prediction method for residual life of aircraft engine | |
| CN109611087B (en) | Volcanic oil reservoir parameter intelligent prediction method and system | |
| CN111458748B (en) | Performance earthquake motion risk analysis method based on three-layer data set neural network | |
| CN112580263B (en) | Turbofan engine residual service life prediction method based on space-time feature fusion | |
| CN111830408A (en) | Motor fault diagnosis system and method based on edge calculation and deep learning | |
| CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
| CN111680875B (en) | Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model | |
| CN111127246A (en) | Intelligent prediction method for transmission line engineering cost | |
| US20220261655A1 (en) | Real-time prediction method for engine emission | |
| CN115270635B (en) | Bayes-neural network high-rise building earthquake demand and vulnerability prediction method | |
| CN111190429B (en) | Unmanned aerial vehicle active fault-tolerant control method based on reinforcement learning | |
| CN113469470B (en) | Energy consumption data and carbon emission correlation analysis method based on electric brain center | |
| CN110378744A (en) | Civil aviaton's frequent flight passenger value category method and system towards incomplete data system | |
| CN115828140A (en) | Neighborhood mutual information and random forest fusion fault detection method, system and application | |
| CN110874616A (en) | Transformer operation prediction method based on LSTM network and Markov chain correction error | |
| CN114266278A (en) | Dual-attention-network-based method for predicting residual service life of equipment | |
| CN114880767A (en) | Aero-engine remaining service life prediction method based on attention mechanism Dense-GRU network | |
| Che et al. | Multi-head self-attention bidirectional gated recurrent unit for end-to-end remaining useful life prediction of mechanical equipment | |
| CN114399064A (en) | Equipment health index construction method based on multi-source sensor data fusion | |
| Yun et al. | Research on gas pressure regulator fault diagnosis based on deep confidence network (DBN) theory | |
| CN110674791B (en) | Forced oscillation layered positioning method based on multi-stage transfer learning | |
| Samuel et al. | Fast modelling of gas reservoir performance with proper orthogonal decomposition based autoencoder and radial basis function non-intrusive reduced order models | |
| CN108760813A (en) | A kind of turbine blade of gas turbine health monitoring systems and method based on temperature signal | |
| CN115169235A (en) | Super surface unit structure inverse design method based on improved generation of countermeasure network | |
| CN111625901B (en) | Intelligent pressure coefficient curve generation method for wing profile |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |