Aero-engine service life prediction method based on knowledge and data fusion driving

By combining knowledge graphs and random differential equations, the complex environment and uncertainty problems in aircraft engine life prediction are solved, and higher prediction accuracy and generalization capabilities are achieved, which significantly improves the performance of the model.

CN118965215AActive Publication Date: 2024-11-15TAIYUAN UNIVERSITY OF TECHNOLOGY

Patent Information

Application Number
CN202410960278.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-17
Publication Date
2024-11-15
Estimated Expiration
2044-07-17

AI Technical Summary

Technical Problem

The existing aero engine life prediction methods are difficult to accurately model when facing complex operating environments and high uncertainties, and lack in-depth understanding of physical characteristics and operating mechanisms, resulting in poor generalization ability and robustness.

Method used

Using a method based on knowledge and data fusion, the data is preprocessed through K-means clustering and exponential weighted moving average method, combined with stochastic differential equations and domain knowledge to build an aircraft engine RUL prediction model, and using GRU and knowledge graph for feature extraction and Relational modeling and improve the interpretability and generalization capabilities of the model through adaptive fusion.

Benefits of technology

It improves the accuracy and generalization ability of aircraft engine life prediction, and can more accurately characterize the uncertainty and randomness of the degradation process. Compared with traditional methods, it achieves a 32% accuracy improvement on the C-MAPSS dataset.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118965215A_ABST
    Figure CN118965215A_ABST
Patent Text Reader

Abstract

The invention provides an aero-engine service life prediction method based on knowledge and data fusion driving, and belongs to the technical field of aero-engine service life prediction. The technical problem to be solved is to provide the aero-engine service life prediction method based on knowledge and data fusion driving. According to the technical scheme, an overall architecture of a coder-decoder is adopted, the coder extracts advanced feature representation from past observation, and the decoder summarizes past information and continuously optimizes a prediction result; a GRU-SDE module is designed in the encoder, a stochastic differential equation SDE is used for simulating the degradation process of the engine, and the uncertainty and randomness of degradation are simulated more accurately while time sequence dependence is extracted; constructing a knowledge graph for the inherent knowledge of the engine, and obtaining a knowledge-based multi-sensor relation matrix; randomly initializing a data-based relation matrix, and realizing adaptive fusion in the training process of the data-based relation matrix and the data-based relation matrix; the method is applied to aero-engine life prediction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention provides an aero-engine life prediction method based on knowledge and data fusion drive, belonging to the technical field of aero-engine life prediction. Background Art

[0002] With the rise and advancement of smart manufacturing and Industry 4.0 concepts, the importance of equipment health management (Prognostics and Health Management, PHM) and predictive maintenance (Predictive Maintenance, PdM) has become increasingly prominent, and has become increasingly important in ensuring the reliability, safety and economic benefits of production systems. In this context, remaining useful life prediction (Remaining Useful Life, RUL) as an important research direction for industrial applications requires an effective estimation of the expected duration of mechanical equipment from the current moment to complete failure.

[0003] As the power component of the aircraft, the reliability and safety of aircraft engines are directly related to the safety of flight. However, during operation, the engine will inevitably be affected by various factors, such as high temperature, high pressure, vibration, corrosion, etc., which will cause its performance to gradually degrade and its remaining service life to continue to decrease. Therefore, accurately predicting the remaining service life of aircraft engines is of great significance to ensuring flight safety, optimizing maintenance strategies and reducing operating costs.

[0004] At present, the main prediction methods for the remaining useful life of aircraft engines are: predicting RUL by deeply understanding the physical degradation process and failure mechanism of the engine. However, due to the high complexity of the system operation mechanism, the cost of obtaining an accurate physical model is high; in addition, deep learning technology has become an effective means to process highly nonlinear and changing sequence data, including fully connected neural networks (FCN), recurrent neural networks (RNN), convolutional neural networks (CNN), Transformer based on attention mechanism, graph neural network (GNN), etc. The above advanced network architecture can be used to obtain better results in RUL prediction tasks than traditional methods.

[0005] The aircraft engine RUL prediction technology based on deep learning can be divided into two categories: methods based on time patterns and methods based on spatiotemporal patterns:

[0006] The first type of strategy focuses on mining the complex temporal changes within the sequence; the classic network architectures include long short-term memory networks (LSTM) and gated recurrent units (GRU), which rely on the Markav assumption to model sequence data. For example, Sayah et al. proposed a multi-layer LSTM for aircraft engine RUL prediction. Luo et al. combined posterior analysis with a large number of GRU classifiers to effectively solve the multi-path degradation problem in life prediction. Although these methods have achieved some good results, their efficiency is affected by the sequential computing paradigm. In contrast, the temporal convolutional network (TCN) has shown unique advantages in the RUL prediction task due to its parallel computing capabilities and efficient feature extraction mechanism. Wang et al. used TCN to process multi-dimensional and multi-channel sensor data and proposed a new temporal convolutional neural network with a soft threshold mechanism for engine RUL prediction. Cao et al. used TCN to capture temporal dependencies and extract high-level feature representations from vibration signals and achieved good prediction results. However, due to the limitations of one-dimensional convolution kernels, they can only model changes between adjacent time points. In recent years, Transformers with attention mechanisms have been widely used to extract long-term dependencies from time series data. Zhang et al. used the attention mechanism to process the association between non-adjacent time points, and proposed an integrated multi-head dual sparse self-attention network based on the improved Transformer to improve the RUL prediction effect.

[0007] The second type of strategy often considers the potential spatial correlation between multiple sensors based on the first type of strategy. Related studies usually use attention mechanisms or graph neural networks to learn the relationship between sensors to enhance the contribution value of features that play a key role in prediction. For example, Liu et al. proposed a new data-driven aircraft engine RUL estimation framework. They used a CNN with channel attention to obtain more important key elements. Subsequently, the Transformer model was used to focus on these important features. In order to reveal the features most closely related to the degradation process, Chen et al. added an attention mechanism to LSTM, which enabled them to give more weight to important samples and features. Sun et al. constructed multiple sensors into a sensor network and used a hierarchical graph representation layer to model spatial dependencies to achieve effective information fusion of sensors.

[0008] However, although the above methods can all predict the service life of aircraft engines, they mainly benefit from the progress in capturing time dependence and spatial correlation from sensor monitoring data, and still have the following limitations and defects:

[0009] (1) In the aviation field, the operating environment of the engine system is complex and changeable, and is affected by a variety of unpredictable factors, such as changes in operating conditions, environmental disturbances, and equipment aging. These factors lead to a high degree of uncertainty in the degradation process of aircraft engines, and its degradation trajectory is obscured by noise, making it difficult to accurately model.

[0010] (2) There are deficiencies in obtaining sensor latent spatial associations based on data. They usually lack a deep understanding of the physical characteristics and operating mechanisms of the engine. This defect limits the generalization ability and robustness of the model when facing changes in data distribution. In fact, domain knowledge has been proven to guide deep learning modeling, but how to combine it with data-driven methods is still a major difficulty. Summary of the invention

[0011] In order to overcome the deficiencies in the prior art, the present invention aims to solve the technical problem of providing an aircraft engine life prediction method based on knowledge and data fusion drive.

[0012] In order to solve the above technical problems, the technical solution adopted by the present invention is: an aircraft engine life prediction method based on knowledge and data fusion drive, including the following prediction steps:

[0013] Step 1: Use the K-means clustering algorithm to obtain different conditional classifications based on the operation settings, and use the z-score operation to standardize the data for each conditional classification category. The calculation formula used for the processing is:

[0014]

[0015] Where: S represents the set of sensors, C represents the set of conditions, i and j represent the i-th sensor and j-th condition classification respectively, u i,j and σ i,j They represent the mean and standard deviation of the ith sensor under the jth condition classification, respectively, and x i,j and z i,j Represent the data before and after standardization, respectively;

[0016] Then the exponentially weighted moving average method is used to smooth the data. The calculation formula used for processing is:

[0017] z t =α*z t +(1-α)*z t-1 ;

[0018] Where: z′ t Indicates z t The filtered value, α represents the filter coefficient, and t represents the time step;

[0019] Step 2: Construct an aircraft engine RUL prediction model based on stochastic differential equations and domain knowledge:

[0020] Step 2-1: Use standard 1×1 convolution to transform the preprocessed data Z∈R T×N Projecting to the potential high-dimensional space, we obtain the embedding representation X of the original data, expressed as: X = Embedding (Z);

[0021] Step 2-2: Use the encoder to extract high-level feature representation of the data:

[0022] Step 2-2-1: Process the data through the gated recurrent unit GRU to capture the temporal dependencies and obtain the hidden state;

[0023] Step 2-2-2: Model the engine degradation process using stochastic differential equations (SDE);

[0024] Step 2-2-3: Use the gated recurrent unit GRU output to solve the stochastic differential equation SDE and obtain the feature matrix of the original data;

[0025] Step 2-2-4: Construct an aircraft engine knowledge graph, and specifically summarize the composition relationship, connection relationship, flow relationship, and monitoring relationship between aircraft engine components;

[0026] Step 2-2-5: Perform knowledge transformation to obtain a knowledge-based multi-sensor correlation matrix;

[0027] Step 2-2-6: perform knowledge embedding and adaptive fusion;

[0028] Step 2-3: Use the linear residual module as the decoder of the model, map the encoded hidden representation to the latent space, map the hidden layer to the output linear layer while using dropout, and use layer normalization at the output. The expression is as follows:

[0029] Y = Decoder (O);

[0030] Step 2-4: The decoder output Y will pass through the linear layer to obtain the final aircraft engine RUL prediction value, expressed as:

[0031] RUL = Linear(Y);

[0032] Step 3: Train and test the network constructed in step 2, select the root mean square error RMSE as the loss function, and use the Adam optimization algorithm with adaptive learning rate to accelerate the convergence speed of the model and improve the stability of training by dynamically adjusting the learning rate of each parameter.

[0033] The specific method of capturing the timing dependency in step 2-2-1 and obtaining the hidden state is:

[0034] The data is processed by the gated recurrent unit GRU to capture the temporal dependencies and obtain the hidden state h t , the expression is:

[0035] r t =σ(x t W xr +h t-1 W hr +b r );

[0036] u t =σ(x t W xu +h t-1 W hu +b u );

[0037] c t =RELU(x t W xc +r t ☉W hc +b c );

[0038] h t =(1-u t )☉h t-1 +u t ⊙c t ;

[0039] Where: ⊙ is the element-by-element product, σ is the sigmoid activation function, x t represents the input of this layer at time t, W xr , W hr , W xu , W hu , W xc , W hc , b r , b u and b u are all model parameters.

[0040] The specific method for modeling the engine degradation process in step 2-2-2 is:

[0041] The expression of the stochastic differential equation SDE used is:

[0042] dX t =u(t,X t )dt+σ(t,X t )dW t , X0=x0;

[0043] Where: X t represents the state of the system at time t; u(t, X t ) and σ(t, X t ) are deterministic functions, representing the drift and diffusion characteristics of the system respectively; dW t is the differential of the Wiener process or Brownian motion, representing the random fluctuations or noise to which the system is subject;

[0044] The Euler-Amaru numerical method is used to solve the problem. By discretizing the time step and gradually approaching the solution of SDE, the evolution trajectory of the system state over time is obtained. The calculation formula is:

[0045]

[0046] Where: Δt is the length of the unit time period; u(t, X t ) and σ(t, X t ) represent drift term and diffusion term respectively; ∈ t represents the noise term;

[0047] According to the above formula, the state X of each step is obtained by iterative solution. t .

[0048] The specific method of obtaining the characteristic matrix of the original data in step 2-2-3 is:

[0049] The hidden state h extracted by the gated recurrent unit GRU t Input into the stochastic differential equation SDE and calculate the drift term u t (t,X t ), diffusion term σ t (t,X t ), noise term∈ t (t,X t ), using three different linear layers to transform h t Project to latent space to obtain unknown item representation;

[0050] For ∈ t , the sigmoid function is used to adaptively adjust the strength of the randomly generated noise z, the expression is:

[0051] u t (t,x t )=Linear u (h t );

[0052] σ t (t,x t )=Linear σ (h t);

[0053] ∈ t (t,x t )=sigmoid(Linear ∈ (h t ))*z t ;

[0054] Where: Linear u , Linear σ , Linear ∈ represents three different linear layers, sigmoid represents the activation function, z t is the noise randomly generated at time t and follows a normal distribution;

[0055] Then the drift term u t (t,X t ), diffusion term σ t (t,X t ), noise term∈ t (t,X t ) into the following formula:

[0056]

[0057] Iterate to solve the state X at each step t ;

[0058] The model is constructed by calculating the difference between adjacent time steps, expressed as:

[0059]

[0060]

[0061] Differential sequence ΔX={Δx1,Δx2...Δx t , ...Δx T} can reflect the changing trend and dynamic characteristics of time series data. Through the above steps, the characteristic matrix X of the original data is obtained. fea .

[0062] The specific method for constructing the aircraft engine knowledge graph in step 2-2-4 is:

[0063] Summarize the composition relationships: describe the components that make up the overall system of the turbofan engine;

[0064] Summarize the connections: Describe the physical connections between the components in a turbofan engine;

[0065] Summarize flow relationships: Describe the flow paths of gases and liquids in a turbofan engine;

[0066] Summarize the monitoring relationships: describe which sensors monitor the key parameters of each component in the turbofan engine;

[0067] Based on the above summarized information, the knowledge graph of turbofan engines is constructed.

[0068] The specific method for knowledge conversion in step 2-2-5 is:

[0069] For a triple (u, v, w) in the knowledge graph, the TransE model is used to learn the vector representation of entities and relations by optimizing the following objective function:

[0070]

[0071] Where: u, v, w represent the vector representation of the head entity, relation and tail entity respectively, and θ represents the set of all triples in the knowledge graph;

[0072] By optimizing the above objective function, the embedding e of each entity is obtained, and then the Pearson correlation coefficient is used to calculate the similarity between entities to obtain the knowledge-based multi-sensor correlation matrix A1.

[0073] The specific method of performing knowledge embedding and adaptive fusion in step 2-2-6 is:

[0074] The data-based multi-sensor correlation matrix A2 is randomly initialized, the knowledge-based correlation matrix A1 and the data-based correlation matrix A2 are fused to generate a comprehensive correlation matrix A. The weights of the two matrices are dynamically adjusted using the adaptive weight allocation method:

[0075] Let A1∈R D×D and A2∈R D×D The expanded dimension is A1∈R 1×D×D and A2∈R 1×D×D , and then use 2D convolution to obtain the adaptive weight parameter α, the expression is:

[0076]

[0077] A=αA1+(1-α)A2;

[0078] Where: σ is the sigmoid activation function, g(·) represents the 2D convolutional layer, represents matrix concatenation; α is an adaptive weight parameter, and its value range is between [0,1];

[0079] By optimizing the value of α, the effective fusion of A1 and A2 is achieved;

[0080] Feature matrix X feaMultiplying the fused sensor relationship matrix A, the encoder output will be represented as o, expressed as:

[0081] O=AX fea .

[0082] The beneficial effects of the present invention compared with the prior art are as follows: the present invention proposes an aircraft engine life prediction method based on knowledge and data fusion, which involves aircraft engine fault diagnosis and life prediction. Specifically, the present aircraft engine degradation process has a high degree of uncertainty and is difficult to accurately model, and the traditional deep learning method has high data dependence and poor generalization, and cannot effectively utilize domain knowledge. The GRU-SDE module is designed, and the stochastic differential equation is innovatively used to simulate the degradation process of the aircraft engine, which more accurately characterizes the uncertainty and randomness of the degradation process; the present invention constructs a knowledge graph for aircraft engine domain knowledge, and combines it with the data-driven learning process through adaptive fusion, effectively embedding the domain knowledge into the deep neural network, and improving the interpretability and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] The present invention will be further described below in conjunction with the accompanying drawings:

[0084] Figure 1 The present invention is a flowchart of the steps of aircraft engine RUL prediction based on stochastic differential equations and domain knowledge;

[0085] Figure 2 It is a structural diagram of the aircraft engine RUL prediction model of the present invention;

[0086] Figure 3 This is a schematic diagram of the present invention using GRU to obtain the unknown term representation of SDE;

[0087] Figure 4 It is a schematic structural diagram of a turbofan engine of the present invention;

[0088] Figure 5 A schematic diagram of the knowledge map of the turbofan engine of the present invention;

[0089] Figure 6 This is a structural diagram of the decoder used in the present invention. DETAILED DESCRIPTION

[0090] The present invention innovatively combines deep learning with domain knowledge to develop a high-precision aircraft engine life prediction model, which not only promotes technological progress in the field of fault diagnosis and life prediction, but also provides new ideas for the health management of other complex industrial systems. With the promotion and application of the method, it can drive the technological upgrading of related industries, improve the intelligence level of equipment, and promote the development of industrial Internet and intelligent manufacturing. The technology proposed in the present invention has been rigorously and comprehensively evaluated on the C-MAPSS turbofan engine data set, and has achieved a 32% improvement in accuracy compared to traditional methods. It is an effective aircraft engine life prediction method. The prediction method mainly includes: adopting the overall architecture of the encoder-decoder, The encoder extracts high-level feature representations from past observations, and the decoder summarizes past information and continuously optimizes the prediction results; a GRU-SDE module is designed in the encoder, and the stochastic differential equation (SDE) is used to simulate the engine degradation process, which can more accurately simulate the uncertainty and randomness of degradation while extracting timing dependencies; a knowledge graph is constructed for the inherent knowledge of the engine to obtain a knowledge-based multi-sensor relationship matrix; at the same time, the data-based relationship matrix is ​​randomly initialized, and the two are adaptively fused during the training process; the decoder uses the residual multi-layer perceptron (MLP) as the core architecture to simply and efficiently map the encoded hidden representation to a high-dimensional space; finally, a linear layer is used to obtain the final aircraft engine RUL prediction.

[0091] In order to more clearly illustrate and explain the purpose, measures and key points of the present invention, the method proposed by the present invention is described in detail below with reference to the accompanying drawings:

[0092] The main purpose of the present invention is to achieve the remaining service life prediction of aircraft engines. Figure 1 The figure shows the overall flow chart of the remaining useful life prediction algorithm of the present invention. Figure 2 The model framework diagram is attached. Figure 3 , Attachment Figure 4 , Attachment Figure 5 , Attachment Figure 6 The SDE unknown item representation diagram, turbofan engine structure and sensor location diagram, turbofan engine knowledge graph and decoder structure diagram are obtained by using GRU respectively; Table 1 shows the 21 measurement sensors inside the turbofan engine and their meanings, and Table 2 shows the meanings of the constructed triples; the following first introduces the research problems of the present invention and the overall algorithm flow proposed, and then describes the design details of each module in detail. The specific implementation method is as follows:

[0093] 1. Problem definition:

[0094] In the field of reliability engineering, the RUL prediction of aircraft engines is a regression problem whose input is a multi-source time series X∈R T×N, where T represents the length of the historical time step, and N represents the number of sensors; the task objective of the present invention is to predict the engine RUL in real time by establishing a mapping relationship between RUL and historical sensor data, and construct a mapping function f, expressed as RUL=f(X).

[0095] 2. Model framework:

[0096] The invention proposes a method for predicting the life of an aircraft engine based on knowledge and data fusion, which involves a degradation process modeling method based on stochastic differential equations and domain knowledge, and is used for predicting the RUL of an aircraft engine. The established model framework is as follows: Figure 2 As shown. The model adopts a standard encoder-decoder structure, in which the encoder extracts information from past observations, and the decoder summarizes past information and improves predictions from simple initialization; specifically, a GRU-SDE module is designed in the encoder, and a stochastic differential equation (SDE) is used to simulate the system degradation process, which can more accurately simulate the uncertainty and randomness of degradation while extracting timing dependencies; in addition, a knowledge graph is constructed for the inherent knowledge of aircraft engines, a multi-sensor relationship matrix is ​​obtained and adaptively fused with data, and a balance between knowledge and data is found during the training process; the decoder used in the present invention uses a residual linear layer, which uses a multi-layer perceptron (MLP) as the core architecture to simply and efficiently map the encoded hidden representation to a high-dimensional space; finally, a linear layer is used to obtain the final RUL prediction.

[0097] The detailed modeling and prediction process is as follows:

[0098] Step 1: Data preprocessing:

[0099] Step 1-1: In order to effectively deal with the substantial impact of the changing working conditions of aircraft engines on sensor data, the K-means clustering algorithm is used to obtain different condition classifications based on the operation settings, and the z-score is used to standardize the data for each category. The calculation formula used for the processing is:

[0100]

[0101] Where: s represents the set of sensors, c represents the set of conditions, i and j represent the i-th sensor and j-th condition classification respectively, u i,j and σ i,j They represent the mean and standard deviation of the ith sensor under the jth condition classification, respectively, and x i,j and z i,j Represent the data before and after normalization, respectively.

[0102] Step 1-2: Use the exponentially weighted moving average method to smooth the data, reduce the impact of local oscillations in the sensor data, and obtain a more stable and reliable sensor data representation. The calculation formula used for processing is:

[0103] z t =α*z t +(1-))*z t-1 ;

[0104] Where: z′ t Indicates z t The filtered value, α represents the filter coefficient, and t represents the time step.

[0105] Step 2: Construct an aircraft engine RUL prediction model based on stochastic differential equations and domain knowledge:

[0106] Step 2-1: Data embedding:

[0107] Using standard 1×1 convolution, the preprocessed data Z∈R T×N Projecting to the potential high-dimensional space, we obtain the embedding representation X of the original data, and the formula is:

[0108] X=Embedding(Z).

[0109] Step 2-2: Use the encoder to extract high-level feature representation of the data:

[0110] Step 2-2-1: Process the data through the Gated Recurrent Unit (GRU), capture the temporal dependencies, and obtain the hidden state h t , the expression is:

[0111] r t =σ(x t W xr +h t-1 W hr +b r );

[0112] u t =σ(x t W xu +h t-1 W hu +b u );

[0113] c t =RELU(x t W xc +r t ☉W hc +b c );

[0114] ht =(1-u t )☉h t-1 +u t ☉c t ;

[0115] Where: ⊙ is the element-by-element product, σ is the sigmoid activation function, x t represents the input of this layer at time t, w xr , W hr , W xu , W hu , W xc , W hc , b r ,b u and b u are all model parameters.

[0116] Step 2-2-2: Use stochastic differential equations to model the engine degradation process. Stochastic differential equations (SDE) are a mathematical tool used to describe the evolution of dynamic systems affected by random forces. The general form of SDE can be expressed as:

[0117] dX t =u(t,X t )dt+σ(t,X t )dW t , X0=x0;

[0118] Where: X t represents the state of the system at time t; u(t, Z t ) and σ(t, X t ) are deterministic functions, representing the drift and diffusion characteristics of the system respectively; dW t is the differential of the Wiener process or Brownian motion, representing the random fluctuations or noise to which the system is subject;

[0119] The solution of SDE is usually difficult to obtain directly, so it needs to be solved with the help of numerical methods. Euler-Amaru is one of the simplest and most intuitive numerical methods. By discretizing the time step, the solution of SDE is gradually approached to obtain the evolution trajectory of the system state over time; after solving with the Euler-Amaru method, the following formula can be obtained:

[0120]

[0121] Where: Δt is the length of the unit time period; u(t, X t ) and σ(t, X t ) represent drift term and diffusion term respectively; ∈ tRepresents the noise term; Using the above formula, the state X of each step can be obtained by iterative solution t .

[0122] Step 2-2-3: Use GRU output to solve SDE. t Input into SDE and calculate the drift term u t (t,X t )Diffusion termσ t (t,X t ) and the noise term ∈ t , these terms can be estimated by neural networks or other function approximators. In the present invention, three different linear layers are used to t Projected into the latent space to obtain the unknown item representation (such as Figure 3 ). In order to effectively adapt to the noise in the data, for ∈ t , the sigmoid function is used to adaptively adjust the strength of the randomly generated noise z;

[0123] u t (t,x t )=Linear u (h t );

[0124] σ t (t,x t )=Linear σ (h t );

[0125] ∈ t (t,x t )=sigmoid(Linear ∈ (h t ))*z t ;

[0126] Where: Linear u ,Linear σ ,Linear ∈ represents three different linear layers, sigmoid represents the activation function, z t is the noise randomly generated at time t and follows a normal distribution;

[0127] Then the drift term u t (t,X t ), diffusion term σ t (t,X t ) and the noise term ∈ t (t,X t )Bring-in The state X of each step can be obtained by iterative solution t. The initial state X0 is crucial for modeling the entire time series. However, in practical problems, X0 is often unknown, and random initialization will have a significant impact on the performance of the model. In addition, the complexity of iterative calculations may significantly increase the computational cost, affecting the efficiency and performance of the model. Inspired by the idea of ​​difference, the present invention constructs a model by calculating the difference between adjacent time steps, which cleverly alleviates the above problems, as shown in the following formula:

[0128]

[0129]

[0130] Differential sequence ΔX={Δx1,Δx2...Δx t , ...Δx T} can reflect the changing trend and dynamic characteristics of time series data. Therefore, through the above steps, the characteristic matrix X of the original data can be obtained fea .X fea The temporal dependencies in the data are extracted, and the random dynamics and perturbations in the degradation process are learned.

[0131] Step 2-2-4: Constructing the knowledge graph of aircraft engines:

[0132] The performance of an aircraft engine depends on its individual components. Sorting out the relationships between aircraft engine components can more clearly utilize the relationships between measurement points and the inherent knowledge of aircraft engines. Figure 4 The structure of the turbofan engine used in the C-MAPSS simulation data set provided by NASA is shown in Table 1. The 21 measurement sensors placed inside the turbofan engine and their meanings are given;

[0133]

[0134] Table 1 21 measurement sensors inside a turbofan engine and their meanings

[0135] The composition relationship, connection relationship, flow relationship and monitoring relationship between engine components are summarized as follows:

[0136] Composition relationship: refers to the various components that make up the overall system of a turbofan engine. For example, (engine, composition, fan) means that the engine is composed of a fan;

[0137] Connection relationship: describes the physical connection between the components in the turbofan engine. These connections ensure that the components can work together to achieve the overall function of the engine. For example, (low-pressure rotor, connection, high-pressure compressor) means that the low-pressure rotor and the high-pressure compressor are connected to each other;

[0138] Flow relationship: describes the flow path of gas and liquid in a turbofan engine. These flow relationships are crucial to the normal operation of the engine and determine the efficiency and stability of energy transfer. For example, (fan, flow, low-pressure compressor) means that air flows from the fan to the low-pressure compressor;

[0139] Monitoring relationship: describes which sensors monitor the key parameters of each component in the turbofan engine. For example, (fan, monitoring, S1) means that the fan is monitored by S1;

[0140] From the above, for the construction of the turbofan engine knowledge graph, 29 entities and 4 relationships are summarized, forming a total of 42 triples, such as Figure 5 And as shown in Table 2.

[0141]

[0142] Table 2 Meaning of the constructed triples

[0143] Step 2-2-5: Conduct knowledge conversion:

[0144] Relying on the algorithm in TransE, it learns a low-dimensional vector for each entity and relationship. Specifically, for a triple (u, v, w) in the knowledge graph, the TransE model learns the vector representation of entities and relations by optimizing the following objective function:

[0145]

[0146] Where: u, v, w represent the vector representation of the head entity, relation and tail entity respectively, and θ represents the set of all triples in the knowledge graph.

[0147] By optimizing the above objective function, the embedding e of each entity can be obtained. Then, the similarity between entities is calculated using the Pearson correlation coefficient to obtain the knowledge-based multi-sensor correlation matrix A1.

[0148] Step 2-2-6: Knowledge embedding and adaptive fusion:

[0149] The data-based multi-sensor correlation matrix A2 is randomly initialized, and the channel dependency is learned during the training process. Adaptive fusion aims to fuse the knowledge-based correlation matrix A1 and the data-based correlation matrix A2 to generate a comprehensive correlation matrix A. The present invention adopts an adaptive weight allocation method to dynamically adjust the weights of the two matrices.

[0150] First, A1∈R D×D and A2∈R D×D The expanded dimension is A1∈R 1×D×D and A2∈R 1×D×D, in order to facilitate matrix concatenation. Then, 2D convolution is used to obtain the adaptive weight parameter α;

[0151]

[0152] A=αA1+(1-α)A2;

[0153] Where: σ is the sigmoid activation function, g(·) represents the 2D convolutional layer, represents matrix concatenation; α is an adaptive weight parameter, and its value range is [0,1].

[0154] By optimizing the value of α, we can achieve effective fusion of A1 and A2. Finally, the feature matrix X fea Multiply it with the fused sensor relationship matrix A to achieve effective fusion of multi-source data. The encoder output will be represented as O, as shown below:

[0155] O=AX fea .

[0156] Step 2-3: Use the linear residual module as the decoder of the model ( Figure 6 ), maps the encoded hidden representation to the latent space. Specifically, the linear residual block is an MLP with a hidden layer activated by ReLU. In addition, residual connections are used to alleviate the gradient vanishing. Finally, dropout is used on the linear layer that maps the hidden layer to the output, and layer normalization is used at the output, as shown below:

[0157] Y = Decoder (O).

[0158] Step 2-4: The decoder output Y will pass through the linear layer to obtain the final RUL prediction value as follows:

[0159] RUL=Linear(Y).

[0160] Step 3: Train and test the network constructed in step 2:

[0161] In order to optimize the performance of the model, the root mean square error (RMSE) is selected as the loss function; the Adam optimization algorithm with adaptive learning rate is adopted to accelerate the convergence speed of the model and improve the stability of training by dynamically adjusting the learning rate of each parameter.

[0162] The method for predicting the remaining useful life of an aircraft engine provided by the present invention adopts the overall framework design of an encoder-decoder, innovatively combines stochastic differential equations with recurrent neural networks, more accurately simulates the uncertainty and randomness in the degradation process, and allows the adjustment of noise intensity to adapt to the dynamics of random degradation; at the same time, it efficiently constructs a domain knowledge graph, calculates the correlation matrix, and dynamically adjusts its weights during the learning process; by incorporating knowledge into the network, the model can accurately understand and reflect complex multi-sensor relationships; and a large number of experiments on the C-MAPSS public data set show that this method has achieved the most advanced level in the RUL prediction task. Compared with the current prediction methods, the present invention has achieved significant improvements in root mean square error (RMSE) and scoring function (scoring).

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the life of an aircraft engine based on knowledge and data fusion, characterized in that: The prediction steps include the following: Step 1: Use the K-means clustering algorithm to obtain different conditional classifications based on the operation settings, and use the z-score operation to standardize the data for each conditional classification category. The calculation formula used for the processing is: Where: S represents the set of sensors, C represents the set of conditions, i and j represent the i-th sensor and j-th condition classification respectively, u i,j and σ i,j They represent the mean and standard deviation of the ith sensor under the jth condition classification, respectively, and x i,j and z i,j Represent the data before and after standardization, respectively; Then the exponentially weighted moving average method is used to smooth the data. The calculation formula used for processing is: With' t =α*z t +(1-α)*z′ t-1 ; Where: z′ t Indicates z t The filtered value, α represents the filter coefficient, and t represents the time step; Step 2: Construct an aircraft engine RUL prediction model based on stochastic differential equations and domain knowledge: Step 2-1: Use standard 1×1 convolution to transform the preprocessed data Z∈R T×N Projecting to the potential high-dimensional space, we obtain the embedding representation X of the original data, expressed as: X = Embedding (Z); Step 2-2: Use the encoder to extract high-level feature representation of the data: Step 2-2-1: Process the data through the gated recurrent unit GRU to capture the temporal dependencies and obtain the hidden state; Step 2-2-2: Model the engine degradation process using stochastic differential equations (SDE); Step 2-2-3: Use the gated recurrent unit GRU output to solve the stochastic differential equation SDE and obtain the feature matrix of the original data; Step 2-2-4: Construct an aircraft engine knowledge graph, and specifically summarize the composition relationship, connection relationship, flow relationship, and monitoring relationship between aircraft engine components; Step 2-2-5: Perform knowledge transformation to obtain a knowledge-based multi-sensor correlation matrix; Step 2-2-6: perform knowledge embedding and adaptive fusion; Step 2-3: Use the linear residual module as the decoder of the model, map the encoded hidden representation to the latent space, map the hidden layer to the output linear layer while using dropout, and use layer normalization at the output, as follows: Y = Decoder (O); Step 2-4: The decoder output Y will pass through the linear layer to obtain the final aircraft engine RUL prediction value, expressed as: RUL = Linear(Y); Step 3: Train and test the network constructed in step 2, select the root mean square error RMSE as the loss function, and use the Adam optimization algorithm with adaptive learning rate to accelerate the convergence speed of the model and improve the stability of training by dynamically adjusting the learning rate of each parameter.

2. The method for predicting the life of an aircraft engine based on knowledge and data fusion drive according to claim 1 is characterized in that: The specific method of capturing the timing dependency in step 2-2-1 and obtaining the hidden state is: The data is processed by the gated recurrent unit GRU to capture the temporal dependencies and obtain the hidden state h t , the expression is: r t =σ(x t W xr +h t-1 W hr +b r ); u t =σ(x t W xu +h t-1 W hu +b u ); c t =RELU(x t IN xc +r t ⊙W hc +b c ); h t =(1-u t )⊙h t-1 +u t ⊙c t ; Where: ⊙ is the element-by-element product, σ is the sigmoid activation function, x t represents the input of this layer at time t, W xr , W hr , W xu , W hu , W xc , W hc , b r , b u and b u are all model parameters.

3. The method for predicting the life of an aircraft engine based on knowledge and data fusion drive according to claim 2 is characterized in that: The specific method for modeling the engine degradation process in step 2-2-2 is: The expression of the stochastic differential equation SDE used is: dX t =u(t,X t )dt+σ(t,X t )dW t ,X0=x0; Where: X t represents the state of the system at time t; u(t, X t ) and σ(t, X t ) are deterministic functions, representing the drift and diffusion characteristics of the system respectively; dW t is the differential of the Wiener process or Brownian motion, representing the random fluctuations or noise to which the system is subject; The Euler-Amaru numerical method is used to solve the problem. By discretizing the time step and gradually approaching the solution of SDE, the evolution trajectory of the system state over time is obtained. The calculation formula is: Where: Δt is the length of the unit time period; u(t, X t ) and σ(t, X t ) represent drift term and diffusion term respectively; ∈ t represents the noise term; According to the above formula, the state X of each step is obtained by iterative solution. t .

4. The method for predicting the life of an aircraft engine based on knowledge and data fusion drive according to claim 3 is characterized by: The specific method of obtaining the characteristic matrix of the original data in step 2-2-3 is: The hidden state h extracted by the gated recurrent unit GRU t Input into the stochastic differential equation SDE and calculate the drift term u t (t,X t ), diffusion term σ t (t,X t ), noise term∈ t (t,X t ), using three different linear layers to transform h t Project to latent space to obtain unknown item representation; For ∈ t , the sigmoid function is used to adaptively adjust the strength of the randomly generated noise z, the expression is: u t (t,x t )=Linear u (h t ); σ t (t,x t )=Linear σ (h t ); ∈ t (t,x t )=sigmoid(Linear ∈ (h t ))*z t ; Where: Linear u , Linear σ ,Linear ∈ represents three different linear layers, sigmoid represents the activation function, z t is the noise randomly generated at time t and follows a normal distribution; Then the drift term u t (t,X t ), diffusion term σ t (t,X t ), noise term∈ t (t,X t ) into the following formula: Iterate to solve the state X at each step t ; The model is constructed by calculating the difference between adjacent time steps, expressed as: Difference sequence ΔX = {Δx1, Δx2…Δxt,…Δx T } can reflect the changing trend and dynamic characteristics of time series data. Through the above steps, the characteristic matrix X of the original data is obtained. fea .

5. The method for predicting the life of an aircraft engine based on knowledge and data fusion drive according to claim 4 is characterized in that: The specific method for constructing the aircraft engine knowledge graph in step 2-2-4 is: Summarize the composition relationships: describe the components that make up the overall system of the turbofan engine; Summarize the connections: Describe the physical connections between the components in a turbofan engine; Summarize flow relationships: Describe the flow paths of gases and liquids in a turbofan engine; Summarize the monitoring relationships: describe which sensors monitor the key parameters of each component in the turbofan engine; Based on the above summarized information, the knowledge graph of turbofan engines is constructed.

6. The method for predicting the life of an aircraft engine based on knowledge and data fusion drive according to claim 5 is characterized by: The specific method for knowledge conversion in step 2-2-5 is: For a triple (u, v, w) in the knowledge graph, the TransE model is used to learn the vector representation of entities and relations by optimizing the following objective function: Where: u, v, w represent the vector representation of the head entity, relation and tail entity respectively, and θ represents the set of all triples in the knowledge graph; By optimizing the above objective function, the embedding e of each entity is obtained, and then the Pearson correlation coefficient is used to calculate the similarity between entities to obtain the knowledge-based multi-sensor correlation matrix A1.

7. The method for predicting the life of an aircraft engine based on knowledge and data fusion drive according to claim 6 is characterized by: The specific method of performing knowledge embedding and adaptive fusion in step 2-2-6 is: The data-based multi-sensor correlation matrix A2 is randomly initialized, the knowledge-based correlation matrix A1 and the data-based correlation matrix A2 are fused to generate a comprehensive correlation matrix A. The weights of the two matrices are dynamically adjusted using the adaptive weight allocation method: Let A1∈R D×D and A2∈R D×D The expanded dimension is A1∈R 1×D×D and A2∈R 1×D×D , and then use 2D convolution to obtain the adaptive weight parameter α, the expression is: A=αA1+(1-α)A2; Where: σ is the sigmoid activation function, g(·) represents the 2D convolutional layer, represents matrix splicing; α is an adaptive weight parameter, and its value range is between [0, 1]; By optimizing the value of α, the effective fusion of A1 and A2 is achieved; Feature matrix X fea Multiplying it with the fused sensor relationship matrix A, the encoder output will be represented as O, expressed as: O = AX fea .

Citation Information

Patent Citations

  • SHA-based spacecraft ACS mission life prediction method

    CN108563142A

  • Equipment residual life prediction method and system considering mutual influence of degradation rates

    CN111460692A

  • Graph structure-based aero-engine remaining service life cross-domain prediction method

    CN116720061A

  • Method for predicting residual life of aero-engine

    CN117113843A

  • Intelligent digital twin simulation system for steel production process

    CN117930786A

Cited By

  • Aero-engine residual life uncertainty quantification method under online prediction scene

    CN119862800A

  • Aero-engine life prediction method based on multi-source time sequence feature fusion

    CN120068651A

  • Aero-engine life prediction method fusing dynamic knowledge graph and dynamic attention mechanism

    CN120124450A

  • Aero-engine health index modeling method fusing linear and nonlinear characteristics

    CN120145671A

  • Method and system for predicting residual service life of aero-engine bearing

    CN120951817A