A method for predicting engine remaining life based on Wiener process and artificial intelligence

Abstract

The present invention relates to a method for predicting the remaining life of an engine based on the Wiener process and artificial intelligence, and belongs to the field of aircraft engine life prediction. In view of the problem that during the operation of a turbofan engine, a variety of degradation modes appear due to different failure behaviors, thereby reducing the prediction accuracy and stability of a single prediction method, a method for identifying different degradation modes of the engine and constructing different models for remaining life prediction is proposed. Prediction is performed according to the following steps: 1) data preprocessing; 2) residual network training; 3) degradation mode classification; 4) prediction model selection and construction; 5) life prediction. The pre-method considering degradation mode recognition can better play the advantages of the corresponding prediction model, thereby improving the effect of life prediction.

Description

Technical Field

[0001] The invention belongs to the field of aircraft engine life prediction and relates to an engine remaining life prediction method based on Wiener process and artificial intelligence. Background Art

[0002] As one of the most important components of an aircraft for space transportation, the aircraft engine is related to the safety and stability of the entire transportation process. The operating environment of an aircraft engine is harsh, with many components and internal structures that affect each other, making it easy for failures to occur. In order to ensure the safe and reliable operation of the equipment, the fault diagnosis and accurate remaining life prediction of the aircraft engine can improve the safety of aircraft operation and promote the development of my country's aviation industry.

[0003] The failure process of aircraft engines is very complex. The degradation process of the engine is affected by many internal and external factors. The fault point has a greater impact on the degradation stage, resulting in the degradation process of the equipment often not just presenting a simple two-stage process, but presenting multiple degradation modes. In this case, the degradation process should be divided into multiple stages. Summary of the invention

[0004] In view of this, an object of the present invention is to provide an engine remaining life prediction method based on Wiener process and artificial intelligence.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for predicting the remaining life of an engine based on Wiener process and artificial intelligence, the method comprising the following steps:

[0007] S1: Data preprocessing: Sliding filter denoising and standardization of engine data;

[0008] S2: Training residual network: Use the training set data to train ResNet101 to divide the engine data set into three categories;

[0009] S3: Degradation mode classification: Input the data set into the residual network in S2 to identify which type of degradation mode the data set belongs to;

[0010] S4: Prediction model selection and construction: According to the degradation mode identified in S3, if it is the first or second type of degradation mode, the ELM-Wiener process prediction model is constructed; if it is the third type of degradation mode, the multi-stage LSTM prediction model is constructed;

[0011] S5: Lifespan prediction: Train the model constructed in S4 and perform lifespan prediction.

[0012] Optionally, S1 specifically includes: using a sliding average method to perform denoising on the original engine detection data, with a sliding window size of 3, and a specific calculation formula is as follows:

[0013]

[0014] Where x(n) is the nth value of the original data, and y(n) is the nth value after smoothing; the denoised data is then subjected to MIN-MAX normalization.

[0015] Optionally, S1 is specifically as follows: firstly, the degradation mode is divided into three categories, namely, a. a type without a fault point or a degradation abnormality; b. a two-stage degradation type containing one fault point; c. a multi-stage degradation type containing two or more fault points; and then a residual network model is trained, the core formula of the residual network model is:

[0016] F(x l ,ω f )=x l +H(x l ,ω l )(2)

[0017] x l is the input of the lth layer, ω l ={ω l,k 1≤k≤K} is the parameter of the lth layer, K is the number of residual unit layers, then the input of the l+1th layer is:

[0018] x l+1 =F(x l ,ω f )(3)

[0019] Substitute this formula into the loop x l+2 =x l+1 +H(x l+1 ,ω l+1 )=H(x l+1 ,ω l+1 )+x l +H(x l ,ω l ),get:

[0020]

[0021] Substituting the data set into the residual network model obtained by S2, it is concluded that the data set belongs to one of the three types of degradation models.

[0022] Optionally, S4 is specifically: if the data set obtained in S3 is a first-class or second-class degradation model, performing the following steps to construct an ELM-Wiener prediction model:

[0023] First, the two-stage Wiener process is used to establish the model, and its mathematical expression is as follows:

[0024]

[0025] Where, μ1 is the drift coefficient of the first stage, μ2 is the drift coefficient of the second stage; σ1 is the diffusion coefficient of the first stage, σ2 is the diffusion coefficient of the second stage; B(t) is the standard Brownian motion; τ is the time of the failure point; x0 represents the initial degradation value of the first stage; x τ is the degradation value at time τ; the mathematical expression of the remaining life is as follows:

[0026] L m = inf{l m :X(t m +l m )≥ωX(t m )≤ω}(6)

[0027] Where L m is the engine at the current time t m The remaining life at time ω is the failure threshold;

[0028] Offline parameter estimation of the prediction model: Let the parameters of the i-th engine be Among them, μ i,1 and μ i,2 are the drift coefficients of the first and second stages, respectively; and are the diffusion coefficients of the first and second stages respectively; τ i is the failure time of the i-th engine;

[0029] Estimate θ using maximum likelihood estimation MLE t , when the function lnL(θ i X i,1:m ) reaches its maximum value, at which point θ i The parameters you want are:

[0030]

[0031] Real-time detection of change points: ELM is used to establish the health index HI. The construction formula of HI is as follows:

[0032] HI 1:k =(hi1,hi2,...,hi k )=Y 1:k -X 1:m (8)

[0033] where Y 1:k =(y1,y2,...,y k) is the output of ELM after training, X 1:k =(X1,X2,…,X k )) is the actual degradation data of the engine;

[0034] Construct HI detection indicators to adaptively detect change points; the form of the detection indicators is as follows:

[0035]

[0036] Parameters online update:

[0037] When t m ≤τ:

[0038]

[0039] where μ 1p:m and They are μ 1p and The posterior estimate of μ 1p,0 and are estimated offline, and are μ 1p and A priori estimate of T 1:m =(t1,t2,...,t m ) is the degenerate data X 1:m The corresponding time, Yes X 1:m The corresponding covariance matrix;

[0040] When t m >τ:

[0041]

[0042] where μ 2p:m and They are μ 2p and The posterior estimate of μ 2p,0 and are obtained by offline estimation, which are μ 2p and A priori estimate of T τ:m =(t τ+1 ,t τ+2 ,...,t τ+m ) is the degenerate data X τ+1:m The corresponding time, Yes X τ+1:m The covariance matrix of

[0043] According to the above parameter solution, substitute it into the life prediction model to predict the remaining life. The model is as follows:

[0044] When t m <τ, that is, the current time t m Before τ:

[0045] S11: Current time t m and the remaining life l m The sum is less than the occurrence time τ of the fault point,

[0046]

[0047] S12: Current time t m and the remaining life m The sum is greater than the time τ of the fault point, that is, t m +l m >τ,

[0048]

[0049] S13: Current time t m is greater than the time when the fault occurs, t m >τ,

[0050]

[0051] If the data set obtained in S3 is a third-type degradation model, perform the following steps to build a multi-segment LSTM prediction model:

[0052] First, determine the number of clusters K by the elbow method. The formula is as follows:

[0053]

[0054] Among them, C i is the i-th cluster, p is C i The sample points in m i It is C i The centroid of C i is the mean of all samples in , SSE is the clustering error of all samples; observe the downward trend of SSE to determine the optimal number of clusters k;

[0055] K-means is used to divide the degradation data into k stages, where for n sample points {x1, x2, ..., x n}, suppose we want to cluster these sample points into k clusters, and the centers of the k clusters are {μ1,μ2,...,μ k}; Define indicator variable γ ij ∈{0,1}, if the i-th sample is clustered into the k-th cluster, then γ ij =1, otherwise γ ij=0; see (17) to (19) for the specific calculation formula:

[0056]

[0057] After obtaining the cluster center, divide the data set into k stages, establish an LSTM model for each stage, and refer to the LSTM algorithm formula to calculate the weight of each gate corresponding to each stage. The specific calculation steps are as follows:

[0058] S21: Reversely calculate the value of the error term σ corresponding to each moment, starting from the current time t, and calculate the error term σ-1 at time t-1:

[0059]

[0060] S22: Calculate weight gradients; calculate the gradient of each weight based on the corresponding error term:

[0061]

[0062] Then iteratively calculate the gradient at each moment to obtain the final gradient. The formula is as follows:

[0063]

[0064] The first-stage data obtained by clustering is used as training data for predicting the first stage; the first-stage and second-stage data are used as training data for predicting the second stage, and the first, second, and third-stage data are used as training data for the third stage through clustering, and so on, to obtain the corresponding model for each stage through training.

[0065] Optionally, S5 specifically comprises: inputting the test set data into the model established in step S4 for life prediction; for the first and second degradation models, describing the remaining life trend of the turbofan engine according to equations (13), (14) and (15); for the third degradation model, using the trained multi-stage LSTM to predict each stage of the new degradation data.

[0066] The beneficial effect of the present invention is that: the present invention aims at the problem that during the operation of a turbofan engine, a variety of degradation modes occur due to different failure behaviors, thereby reducing the prediction accuracy and stability of a single prediction method. First, the residual network RstNet101 is trained with training set data, and different degradation modes are identified through the trained model. The prediction method considering degradation mode recognition can better play the advantages of the corresponding prediction model, thereby improving the effect of remaining life prediction.

[0067] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0069] Figure 1 It is a structural framework diagram of the present invention;

[0070] Figure 2 This is the sliding average filtering denoising result of the data set;

[0071] Figure 3 The life prediction flow chart for the ELM-Wiener process;

[0072] Figure 4 This is a multi-stage LSTM lifespan prediction flowchart;

[0073] Figure 5 This is the result diagram of the clustering algorithm;

[0074] Figure 6 This is the prediction result graph of the multi-stage LSTM model. DETAILED DESCRIPTION

[0075] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0076] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0077] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0078] like Figure 1 As shown, the implementation process of the present invention includes:

[0079] 1) Data preprocessing: Perform sliding filtering and denoising on the engine data, and then perform standardization.

[0080] 2) Training residual network: Use the training set data to train ResNet101 to divide the engine data set into three categories.

[0081] 3) Degradation mode classification: Input the data set into the residual network in 2) to identify which type of degradation mode the data set belongs to.

[0082] 4) Prediction model selection and construction: According to the degradation mode identified in 3), if it is the first or second type of degradation mode, an ELM-Wiener process prediction model is constructed; if it is the third type of degradation mode, a multi-stage LSTM prediction model is constructed.

[0083] 5) Lifespan prediction: Train the model constructed in 4) and perform lifespan prediction.

[0084] experiment

[0085] This example combines the performance degradation simulation experiment data set of the turbofan engine disclosed by NASA Prediction Center as the original degradation data to predict its remaining life. Each row of data of each engine sub-data set in this data set contains 26 columns, which respectively correspond to the 26 data measured by the numbered engine in a certain time period. Each column represents a different meaning. For example, the first column of data represents the serial number of the engine, the second column of data represents the number of cycles of the engine, and the third to fifth columns are the operating setting parameters of the engine, which represent the altitude, Mach number and relative temperature respectively. The present invention selects the ninth sensor coefficient: Nc, that is, the actual speed of the core engine, and takes the data of engine No. 17 as an example to predict the remaining life of the turbofan engine.

[0086] 1) Data preprocessing includes two steps. The first step is to use the sliding average method to denoise the engine data set. The core idea is to take the arithmetic average of the observations in a certain n regions around the observation value. The arithmetic mean obtained can be used to replace the observation value. n is the size of the sliding window, which is 3 here, that is,

[0087]

[0088] Where x(n) is the nth value of the original data, and y(n) is the nth value after smoothing.

[0089] The second step is data standardization, which can correct things of different scales and specifications according to a unified standard, thereby eliminating the difference in the impact of different dimensions on the results. The specific formula is:

[0090]

[0091] By rescaling the values ​​of each dimension of the data, the final data vector falls within the interval [0,1]. Figure 2 The figure shows the result of data preprocessing of dataset FD003.

[0092] 2) Training the residual network First, the degradation mode is divided into three categories, namely a. No fault point or abnormal degradation type; b. Two-stage degradation type with one fault point; c. Multi-stage degradation type with two or more fault points. This avoids the mutual interference between degradation data of different degradation modes. Therefore, it is particularly important to use the residual network to accurately classify the model. Train the residual network model, the core formula of the model is:

[0093] F(x l ,ω f )=x l +H(x l ,ω l )(3)

[0094] x l is the input of the lth layer, ω l ={ω l,k 1≤k≤K} is the parameter of the lth layer, K is the number of residual unit layers, then the input of the l+1th layer is:

[0095] x l+1 =F(x l ,ω f )(4)

[0096] Substitute this formula into the loop x l+2 =x l+1 +H(x l+1 ,ω l+1 )=H(xl+1 ,ω l+1 )+x l +H(x l ,ω l ),get:

[0097]

[0098] By bringing the data set into the residual network model obtained in step 2), it can be concluded that the data set belongs to the third type of model among the three types of degradation models. Therefore, K-means clustering is used to divide the data into multiple stages, and LSTM is used to learn the degradation characteristics of each stage according to the corresponding training strategy to perform multi-stage prediction. Figure 3 The life prediction flow chart for the ELM-Wiener process; Figure 4 Flowchart for multi-stage LSTM lifespan prediction.

[0099] First, determine the number of clusters K by the elbow method. The formula is as follows:

[0100]

[0101] Among them, C i is the i-th cluster, p is C i The sample points in m i It is C i The centroid of C i is the mean of all samples in , SSE is the clustering error of all samples; observe the downward trend of SSE to determine the optimal number of clusters k.

[0102] K-means is used to divide the degradation data into k stages, where for n sample points {x1, x2, ..., x n}, suppose we want to cluster these sample points into k clusters, and the centers of the k clusters are {μ1,μ2,...,μ k}; Define indicator variable γ ij ∈{0,1}, if the i-th sample is clustered into the k-th cluster, then γ ij =1, otherwise γ ij =0; see (7) to (9) for the specific calculation formula:

[0103]

[0104]

[0105]

[0106] According to the above algorithm, the clustering stage diagram is as follows Figure 5As shown, we can get a total of 4 cluster centers and 5 stages. We build an LSTM model for each stage and calculate the weight of each gate corresponding to each stage by referring to the LSTM algorithm formula. The specific calculation steps are as follows:

[0107] a. Reversely calculate the value of the error term σ corresponding to each moment, starting from the current time t, and calculate the error term σ-1 at time t-1:

[0108]

[0109] b. Calculate the weight gradient. Calculate the gradient of each weight based on the corresponding error term:

[0110]

[0111] Then iteratively calculate the gradient at each moment to obtain the final gradient. The formula is as follows:

[0112]

[0113] The first-stage data obtained by clustering is used as the training data for predicting the first stage; the first-stage and second-stage data are used as the training data for predicting the second stage, and the first, second, and third-stage data are used as the training data for the third stage through clustering, and so on, and the corresponding model for each stage is obtained through training. After obtaining the model for each stage, the prediction set data is brought into the model, and the final LSTM prediction of the remaining life RMSE is 2.15. The prediction results are as follows Figure 6 shown.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the remaining life of an engine based on Wiener process and artificial intelligence, characterized in that: The method comprises the following steps: S1: Data preprocessing: Sliding filter denoising and standardization of engine data; S2: Training residual network: Use the training set data to train ResNet101 to divide the engine data set into three categories; S3: Degradation mode classification: Input the data set into the residual network in S2 to identify which type of degradation mode the data set belongs to; S4: Prediction model selection and construction: According to the degradation mode identified in S3, if it is the first or second type of degradation mode, the ELM-Wiener process prediction model is constructed; if it is the third type of degradation mode, the multi-stage LSTM prediction model is constructed; S5: Lifespan prediction: train the model constructed in S4 and perform lifespan prediction; The S4 is specifically: if the data set obtained in S3 is a first-class or second-class degradation model, the following steps are performed to construct the ELM-Wiener prediction model: First, the two-stage Wiener process is used to establish the model, and its mathematical expression is as follows: Where, μ1 is the drift coefficient of the first stage, μ2 is the drift coefficient of the second stage; σ1 is the diffusion coefficient of the first stage, σ2 is the diffusion coefficient of the second stage; B(t) is the standard Brownian motion; τ is the time of the failure point; x0 represents the initial degradation value of the first stage; x τ is the degradation value at time τ; the mathematical expression of the remaining life is as follows: L m inf{l m :X(t m +l m )≥ωX(t m )≤ω} (6) Where L m is the engine at the current time t m The remaining life at time ω is the failure threshold; Offline parameter estimation of the prediction model: Let the parameters of the i-th engine be Among them, μ i,1 and μ i,2 are the drift coefficients of the first and second stages, respectively; and are the diffusion coefficients of the first and second stages respectively; τ i is the failure time of the i-th engine; Estimate θ using maximum likelihood estimation MLE t , when the function lnL(θ i |X i,1:m ) reaches its maximum value, at which point θ i The parameters you want are: Real-time detection of change points: ELM is used to establish the health index HI. The construction formula of HI is as follows: HI 1:k =(hi1,hi2,...,hi k )=|Y 1:k -X 1:m | (8) where Y 1:k =(y1,y2,…,y k ) is the output of ELM after training, X 1:k =(X1,X2,...,X k )) is the actual degradation data of the engine; Construct HI detection indicators to adaptively detect change points; the form of the detection indicators is as follows: Parameters online update: When t m ≤τ: where μ 1p:m and They are μ 1p and The posterior estimate of μ 1p,0 and are estimated offline, and are μ 1p and A priori estimate of T 1:m =(t1,t2,...,t m ) is the degenerate data X 1:m The corresponding time, Yes X 1:m The corresponding covariance matrix; When t m > τ: where μ 2p:m and They are μ 2p and The posterior estimate of 2p,0 and are obtained by offline estimation, which are μ 2p and A priori estimate of T τ:m =(t τ+1 ,t τ+2 ,...,t τ+m ) is the degenerate data X τ+1:m The corresponding time, Yes X τ+1:m The covariance matrix of According to the above parameter solution, substitute it into the life prediction model to predict the remaining life. The model is as follows: When t m <τ, that is, the current time t m Before τ: S11: Current time t m and the remaining life l m The sum is less than the occurrence time τ of the fault point, S12: Current time t m and the remaining life l m The sum is greater than the time τ of the fault point, that is, t m +l m >τ, S13: Current time t m is greater than the time when the fault occurs, t m >τ, If the data set obtained in S3 is a third-type degradation model, perform the following steps to build a multi-segment LSTM prediction model: First, determine the number of clusters K by the elbow method. The formula is as follows: Among them, C i is the i-th cluster, p is C i The sample points in m i It is C i The centroid of C i is the mean of all samples in , SSE is the clustering error of all samples; observe the downward trend of SSE to determine the optimal number of clusters k; K-means is used to divide the degradation data into k stages, where for n sample points {x1, x2, ..., x n }, suppose we want to cluster these sample points into k clusters, and the centers of the k clusters are {μ1,μ2,...,μ k }; Define indicator variable γ ij ∈{0,1}, if the i-th sample is clustered into the k-th cluster, then γ ij =1, otherwise γ ij =0; see (17) to (19) for the specific calculation formula: After obtaining the cluster center, divide the data set into k stages, establish an LSTM model for each stage, and refer to the LSTM algorithm formula to calculate the weight of each gate corresponding to each stage. The specific calculation steps are as follows: S21: Reversely calculate the value of the error term σ corresponding to each moment, starting from the current time t, and calculate the error term σ-1 at time t-1: S22: Calculate weight gradients; calculate the gradient of each weight based on the corresponding error term: Then iteratively calculate the gradient at each moment to obtain the final gradient. The formula is as follows: The first-stage data obtained by clustering is used as training data for predicting the first stage; the first-stage and second-stage data are used as training data for predicting the second stage, and the first, second, and third-stage data are used as training data for the third stage through clustering, and so on, to obtain the corresponding model for each stage through training.

2. The method for predicting the remaining life of an engine based on the Wiener process and artificial intelligence according to claim 1, characterized in that: Specifically, S1 is: using the sliding average method to perform denoising on the original engine detection data, with a sliding window size of 3, and the specific calculation formula is as follows: Where x(n) is the nth value of the original data, and y(n) is the nth value after smoothing; the denoised data is then subjected to MIN-MAX normalization.

3. The method for predicting the remaining life of an engine based on the Wiener process and artificial intelligence according to claim 2, characterized in that: Specifically, S1 is as follows: first, the degradation mode is divided into three categories, namely, a. a type without fault points or abnormal degradation; b. a two-stage degradation type with one fault point; c. a multi-stage degradation type with two or more fault points; and then a residual network model is trained. The core formula of the residual network model is: F(x l ,ω f )=x l +H(x l ,ω l ) (2) x l is the input of the lth layer, ω l ={ω l,k |1≤k≤K|} is the parameter of the lth layer, K is the number of residual unit layers, then the input of the l+1th layer is: x l+1 =F(x l ,ω f ) (3) Substitute this formula into the loop x l+2 =x l+1 +H(x l+1 ,ω l+1 )=H(x l+1 ,ω l+1 )+x l +H(x l ,ω l ),get: Substituting the data set into the residual network model obtained by S2, it is concluded that the data set belongs to one of the three types of degradation models.

4. The method for predicting the remaining life of an engine based on the Wiener process and artificial intelligence according to claim 1, characterized in that: The step S5 specifically includes: inputting the test set data into the model established in step S4 for life prediction; for the first and second degradation models, describing the remaining life trend of the turbofan engine according to formula (13), formula (14) and formula (15); for the third degradation model, using the trained multi-stage LSTM to predict each stage of the new degradation data.

Images