Aero-engine life prediction method and device based on Dlinear-cnn model

Abstract

The application relates to an aero-engine life prediction method and device based on a Dlinear-CNN model, and the method comprises the following steps: obtaining an original data set of aero-engine degradation, performing data preprocessing on the original data set to obtain processed data, and dividing the processed data into a training set and a test set; training a pre-constructed Dlinear-CNN model based on the training set, and taking the trained Dlinear-CNN model as an engine life prediction model; inputting the test set into the engine life prediction model after being segmented by a sliding window for testing, using a Hyperband algorithm to optimize hyperparameters to obtain an optimal hyperparameter combination, updating the engine life prediction model with the optimal hyperparameter combination to obtain an optimized engine life prediction model; and predicting the life of a target aero-engine by using the optimized engine life prediction model. The application combines a CNN convolutional neural network to compensate for the poor performance of a Dlinear model used alone to extract the degradation information of the engine, and the problems of slow model convergence speed and large gap between the prediction result and the actual value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine life prediction, and particularly relates to an aero-engine life prediction method and device based on a Dlinear-CNN model. BACKGROUND

[0002] The aero-engine is the heart of the airplane, and is crucial to ensure flight safety, optimize maintenance plans and reduce operating costs. Accurate prediction can help to discover potential failures in advance, avoid in-flight shutdowns or accidents, and reasonably arrange maintenance time to reduce unnecessary maintenance and extend the service life of the engine.

[0003] At present, data-driven prediction methods have broad application prospects in the field of engineering systems. Data-driven models use big data collected by Internet of Things technology, combined with advanced algorithms and machine learning techniques, to monitor and analyze the health status of equipment in real time, and have advantages such as real-time, accuracy and adaptability, and are widely used at present.

[0004] Although the data-based method does not need to deeply understand the physical mechanism and failure principle of the equipment or component, it can handle complex systems, has strong adaptability, and has high prediction accuracy. However, the data-based method also has strong data dependence, and if the data has noise, outliers or missing values, it will seriously affect the accuracy and reliability of the prediction results; the model has poor interpretability, it is difficult to intuitively understand how the prediction model arrives at the result, which brings certain difficulties to subsequent decision-making and maintenance; the real-time performance and computational efficiency are low, and the real-time processing of a large amount of data and timely prediction have low work efficiency; it may be affected by data bias, the historical data used to train the prediction model has bias or insufficient representation, which may lead to poor prediction results in actual application.

[0005] The existing prediction methods are mostly divided into the following,

[0006] (1) Experience-based method: only the historical failure information of the equipment is needed to predict the remaining life of the equipment through experience.

[0007] (2) Model-based method: the corresponding characteristics of the predicted equipment are needed, and an accurate mathematical model is established to obtain the predicted failure time and the corresponding remaining life.

[0008] (3) Data-driven method: the internal real-time data of the aero-engine is collected, which is different from the model-based method, and the data-driven method effectively reveals the relationship between the sensor data and the system degradation degree.

[0009] The shortcomings of the prior art are as follows,

[0010] (1) Experience-based approach: In real-world scenarios, researchers often lack sufficient experience to fully understand the equipment, resulting in relatively low accuracy and a high error rate.

[0011] (2) Model-based methods: Model-based methods require deep knowledge and high-quality data, as well as a sufficient understanding of the model and its mechanisms. With the rapid development of the industrial sector, mechanical systems are becoming increasingly complex, and the intricate interactions between components make it impractical to build comprehensive and accurate models. Model-based design methods lack sufficient flexibility and portability, making it difficult to meet the needs of modern mechanical system design and ensure the efficiency, flexibility, and maintainability of the design. When the equipment structure is too complex, the modeling difficulty is high, which greatly limits the application of this method in the aerospace field.

[0012] (3) Data-driven approach: It is highly dependent on data. If there are problems such as noise, outliers or missing values ​​in the data, it will seriously affect the accuracy and reliability of the prediction results; poor model interpretability: It is difficult to intuitively understand how the prediction model arrives at the results, which brings certain difficulties to subsequent decision-making and maintenance; low real-time performance and computational efficiency: The efficiency of processing large amounts of data in real time and making predictions in a timely manner is low; it may be affected by data bias: The historical data used to train the prediction model may be biased or insufficiently representative, which may lead to poor performance of the prediction results in practical applications. Summary of the Invention

[0013] The purpose of this invention is to at least address one of the shortcomings of the prior art by providing a method and apparatus for predicting the lifespan of aero-engines based on a Dlinear-CNN model.

[0014] To achieve the above objectives, the present invention adopts the following technical solution:

[0015] Specifically, a method for predicting the lifespan of aero-engines based on the Dlinear-CNN model is proposed, including the following:

[0016] Step 1: Obtain the original dataset of aircraft engine degradation, and perform data preprocessing on the original dataset to obtain processed data. Divide the processed data into training set and test set as sample data.

[0017] Step 2: Train the pre-built Dlinear-CNN model based on the training set, and use the trained Dlinear-CNN model as the engine life prediction model;

[0018] Step 3: After the test set is segmented by sliding window, it is input into the engine life prediction model for testing. During the testing process, the Hyperband algorithm is used to optimize the hyperparameters to obtain the optimal hyperparameter combination. Finally, the predicted RUL value is obtained. The RMSE and Score evaluation indicators are used to evaluate the predicted RUL value and evaluate the prediction effect. The engine life prediction model that meets the preset standard evaluation result is used as the optimized engine life prediction model.

[0019] Step 4: Use the optimized engine life prediction model to predict the life of the target aero-engine.

[0020] Furthermore, specifically, step 1 includes the following sub-steps:

[0021] Step 1.1: Set RUL labels for the initial sensor data in the original dataset to obtain processed data, and use a piecewise linear function to label the engine degradation mode;

[0022] Step 1.2: Filter the processed data and remove data that is obviously irrelevant to the engine to obtain the filtered data;

[0023] Step 1.3: Analyze the correlation between each sensor in the filtered data, filter and optimize the sensor data, and remove redundant sensor data to obtain deredundant data.

[0024] Step 1.4: Perform data normalization on the deredundant data to obtain normalized data, so that the processed data conforms to a standard normal distribution, that is, the mean is 0 and the standard deviation is 1. The data normalization is shown in equation (1).

[0025]

[0026] In the formula, x * Let μ and σ represent the standardized feature values, where μ and σ are the mean and standard deviation of feature x, respectively.

[0027] Step 1.5: Remove outliers from the normalized data. Then, divide the processed data into training and test sets as sample data. The outlier detection uses the Z-Score method, as shown in equation (2), which measures the distance of a data point from the mean and compares it with the standard deviation.

[0028]

[0029] In the formula, mean is the mean of dataset X, and SD is the standard deviation of dataset X.

[0030] Furthermore, specifically, the labeled dataset is expanded using data augmentation strategies to obtain a training dataset, including:

[0031] The labeled dataset is augmented using image enhancement techniques to obtain a training dataset. These image enhancement techniques include rotation, cropping, changing the hue, saturation, and brightness of the image, and adding noise.

[0032] Furthermore, specifically, step 2 includes the following sub-steps:

[0033] Step 2.1: Construct the Dlinear model;

[0034] Step 2.2: Perform convolution on the input data to extract features from the data, select features, filter information, and combine features to obtain the final feature vector;

[0035] Step 2.3: Integrate the DCNN layer into the Dlinear model to capture local features and sequence patterns. Set hyperparameters such as kernel size, learning rate, and sliding window length to train the model and obtain the trained Dlinear-CNN model.

[0036] Furthermore, specifically, the sliding window segmentation in step 3 includes,

[0037] The high-dimensional vector X obtained after preprocessing the test set data is divided into trend sequences X using a moving average filter. t and the remaining sequence X s H is then obtained by passing it through two linear layers and a convolutional layer respectively. t H s The summation yields the final prediction result Y, which is expressed as follows:

[0038]

[0039] in:

[0040]

[0041] X s [i] = X[i] - X t [i] (5)

[0042] In the formula, Y represents the predicted remaining engine life, and W... t W t2 W s W s2 K represents the weights of the linear layer. t ,K s For the convolutional layer kernel, b t1 ,b s1 ,b tc ,b sc ,b t2 ,b s2For the bias term, Ds is the DKernalSize hyperparameter of DLinear.

[0043] Furthermore, specifically, step 3, which uses the Hyperband algorithm to optimize hyperparameters and obtain the optimal hyperparameter combination, includes the following:

[0044] Determine the hyperparameters to be tuned, their ranges and step sizes, and the configuration for Hyperband hyperparameter tuning;

[0045] Calculate the number of outer loops Where R is the maximum number of iteration resources that can be allocated to a single hyperparameter combination, η is the downsampling rate, indicating that the best 1 / η hyperparameter combination evaluated after each iteration can enter the next iteration, and the total number of resources B = (s max +1)R, each s corresponds to one outer loop round, s∈{s max ,s max `-1,...,0}` represents the number of hyperparameter combinations for the external circulation cycle in this round. The actual maximum number of resources in the outer circulation of this round is r = Rη -s Obtain n sets of hyperparameter combinations from the hyperparameter space and store them in a sequence T. Each outer loop contains a nested inner loop, with each i corresponding to one inner loop, where i∈{0,1,2,...,s}. For each inner loop, calculate the number of hyperparameter combinations to be evaluated. And the amount of resources r to be allocated in this round of the inner loop. i =rη i For each combination of hyperparameters in T, assign r i Computational resources are used for model training and testing, resulting in test loss. The models with the smallest loss in time T are selected. The parameters are carried over to the next iteration until all iterations are completed, resulting in the optimal set of hyperparameters.

[0046] Furthermore, the method also includes,

[0047] The optimized engine life prediction model was used to predict the RUL value of the dataset. The predicted RUL value was evaluated using RMSE and Score evaluation metrics, and the prediction effect was evaluated.

[0048] This invention also proposes a device for predicting the lifespan of aero-engines based on a Dlinear-CNN model, comprising the following:

[0049] The data acquisition module is used to acquire the original dataset of aircraft engine degradation, and to preprocess the original dataset to obtain processed data. The processed data is then used as sample data to divide the dataset into training set and test set.

[0050] The model training module is used to train a pre-built Dlinear-CNN model based on the training set, and to use the trained Dlinear-CNN model as an engine life prediction model.

[0051] The model testing module inputs the test set into the engine life prediction model after it is segmented by a sliding window. During the testing process, the Hyperband algorithm is used to optimize the hyperparameters to obtain the optimal hyperparameter combination. Finally, the predicted RUL value is obtained. The predicted RUL value is evaluated using RMSE and Score evaluation metrics, and the prediction effect is evaluated. The engine life prediction model that meets the preset standard evaluation result is taken as the optimized engine life prediction model.

[0052] The model deployment module is used to predict the lifespan of a target aero-engine using an optimized engine lifespan prediction model.

[0053] The beneficial effects of this invention are as follows:

[0054] This invention proposes a method and apparatus for predicting the lifespan of aero-engines based on a Dlinear-CNN model. It introduces a Dlinear model to extract key information from sensor data and combines it with CNN to enhance feature extraction, accurately capturing engine degradation characteristics. The method consists of four main steps: data preprocessing, constructing a Dlinear model, optimizing model training data, and evaluating prediction performance. Data preprocessing ensures the validity of the input data. The construction of the Dlinear model divides the input time series into trend and residual parts to analyze long-term trends and short-term changes respectively. For these two parts, Dlinear models utilize independent single-layer linear networks, achieving both efficiency and effectiveness in time series processing. The fusion of the CNN convolutional neural network compensates for the shortcomings of using the Dlinear model alone in extracting engine degradation information, which suffers from slow model convergence and large discrepancies between predicted and actual values. Attached Figure Description

[0055] The above and other features of this disclosure will become more apparent from the detailed description of the embodiments illustrated in conjunction with the accompanying drawings. In the accompanying drawings, the same reference numerals denote the same or similar elements. Obviously, the drawings described below are merely some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort. In the drawings:

[0056] Figure 1 The flowchart shown is a process for predicting the lifespan of an aero-engine based on a Dlinear-CNN model according to the present invention.

[0057] Figure 2The image shown is a comparison chart of the predicted and actual values ​​of the FD001 dataset when this invention is applied.

[0058] Figure 3 The flowchart shown is a process for the hyperparameter optimization algorithm of this invention. Detailed Implementation

[0059] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The same reference numerals used throughout the accompanying drawings indicate the same or similar parts.

[0060] Example 1, referring to Figure 1 This invention proposes a method for predicting the lifespan of aero-engines based on a Dlinear-CNN model, comprising the following:

[0061] Step 1: Obtain the original dataset of aircraft engine degradation, and perform data preprocessing on the original dataset to obtain processed data. Divide the processed data into training set and test set as sample data.

[0062] Step 2: Train the pre-built Dlinear-CNN model based on the training set, and use the trained Dlinear-CNN model as the engine life prediction model;

[0063] Step 3: After the test set is segmented by sliding window, it is input into the engine life prediction model for testing. During the testing process, the Hyperband algorithm is used to optimize the hyperparameters to obtain the optimal hyperparameter combination. Finally, the predicted RUL value is obtained. The RMSE and Score evaluation indicators are used to evaluate the predicted RUL value and evaluate the prediction effect. The engine life prediction model that meets the preset standard evaluation result is used as the optimized engine life prediction model.

[0064] Step 4: Use the optimized engine life prediction model to predict the life of the target aero-engine.

[0065] In a preferred embodiment of the present invention, step 1 specifically includes the following sub-steps.

[0066] Step 1.1: Set RUL labels for the initial sensor data in the original dataset to obtain processed data, and use a piecewise linear function to label the engine degradation mode;

[0067] Step 1.2: Filter the processed data, removing data that is obviously irrelevant to the engine to obtain the filtered data; perform feature selection, extraction and analysis on the engine monitoring data, and filter out data with monotonicity and strong correlation to improve prediction accuracy. Among them, there are sensor data with variance of 0, which means that their monitoring values ​​remain constant as the engine degrades. Therefore, not all sensor data contains engine degradation information. It is necessary to filter the sensor data and remove data that is obviously irrelevant to the engine.

[0068] Step 1.3: Analyze the correlation between each sensor in the filtered data, filter and optimize the sensor data, and remove redundant sensor data to obtain deredundant data.

[0069] Step 1.4: Perform data normalization on the deredundant data to obtain normalized data, so that the processed data conforms to a standard normal distribution, that is, the mean is 0 and the standard deviation is 1. The data normalization is shown in equation (1).

[0070]

[0071] In the formula, x * Let μ and σ represent the standardized feature values, where μ and σ are the mean and standard deviation of feature x, respectively.

[0072] Step 1.5: Remove outliers from the normalized data. Then, divide the processed data into training and test sets as sample data. The outlier detection uses the Z-Score method, as shown in equation (2), which measures the distance of a data point from the mean and compares it with the standard deviation.

[0073]

[0074] In the formula, mean is the mean of dataset X, and SD is the standard deviation of dataset X.

[0075] In a preferred embodiment of the present invention, specifically, the labeled dataset is expanded using data augmentation strategies to obtain a training dataset, including,

[0076] The labeled dataset is augmented using image enhancement techniques to obtain a training dataset. These image enhancement techniques include rotation, cropping, changing the hue, saturation, and brightness of the image, and adding noise.

[0077] In a preferred embodiment of the present invention, step 2 specifically includes the following sub-steps.

[0078] Step 2.1: Construct the Dlinear model;

[0079] Step 2.2: Perform convolution on the input data to extract features from the data, select features, filter information, and combine features to obtain the final feature vector;

[0080] Step 2.3: Integrate the DCNN layer into the Dlinear model to capture local features and sequence patterns. Set hyperparameters such as kernel size, learning rate, and sliding window length to train the model and obtain the trained Dlinear-CNN model.

[0081] In a preferred embodiment of the present invention, specifically, the sliding window segmentation in step 3 includes:

[0082] The high-dimensional vector X obtained after preprocessing the test set data is divided into trend sequences X using a moving average filter. t and the remaining sequence X s H is then obtained by passing it through two linear layers and a convolutional layer respectively. t H s The summation yields the final prediction result Y, which is expressed as follows:

[0083]

[0084] in:

[0085]

[0086] X s [i] = X[i] - X t [i] (5)

[0087] In the formula, Y represents the predicted remaining engine life, and W... t W t2 W s W s2 K represents the weights of the linear layer. t ,K s For the convolutional layer kernel, b t1 ,b s1 ,b tc ,b sc ,b t2 ,b s2 For the bias term, Ds is the DKernalSize hyperparameter of DLinear.

[0088] Reference Figure 3 In a preferred embodiment of the present invention, specifically, step 3, which uses the Hyperband algorithm to optimize hyperparameters and obtain the optimal hyperparameter combination, specifically includes:

[0089] Determine the hyperparameters to be tuned, their ranges and step sizes, and the configuration for Hyperband hyperparameter tuning;

[0090] Calculate outer loop variable Total resources B = (s max +1)R, each s corresponds to one outer loop round, s∈{s max ,s max `-1,...,0}` represents the number of hyperparameter combinations for the external circulation cycle in this round. The actual maximum number of resources in the outer circulation of this round is r = Rη -s Obtain n sets of hyperparameter combinations from the hyperparameter space and store them in a sequence T. Each round of the outer loop contains a nested inner loop, with each i corresponding to one inner loop, where i∈{0,1,2,...,s}. Calculate the number of remaining hyperparameter combinations in the inner loop of that round. Calculate the amount of resources r that need to be allocated in this round of the loop. i =rη i For each combination of hyperparameters in T, assign r i Computational resources (here, computational resources are measured in iterations) are used for model training and testing. The test loss is obtained, and the models with the smallest loss in time interval T are selected. The parameters are carried over to the next iteration until all iterations are completed, resulting in the optimal set of hyperparameters.

[0091] The core idea of ​​Hyperband is to design multiple random sampling frames, randomly generate n hyperparameter combinations within each frame, and assign r to each combination. i Iterate through the resources, and finally retain only the best 1 / η portion of the evaluation result for the next iteration, where η is the downsampling factor, typically set to 3 or 4 by default. The final hyperparameter combination will be allocated the maximum iteration resource R. The specific steps are as follows:

[0092] (1) Input the maximum iterative resource R and the retention probability 1 / η;

[0093] (2) Initialize the number of sampling frames The total resources within the framework, B = (s max +1)R;

[0094] (3) For any frame s∈{s max ,s max Execute: -1,…,0}

[0095] 1) Calculate the number of hyperparameter combinations for random sampling. And the iterative resources r = Rη for each hyperparameter combination -s ;

[0096] 2) Randomly sample n hyperparameter combinations within the hyperparameter definition space and store them all in T;

[0097] 3) For any i∈{0,…,s}, the continuous downsampling process is performed as follows:

[0098] ① Initialize the number of combinations to be evaluated and its iterative resource r i =rη i ;

[0099] ② Assign r to each hyperparameter combination in T i Iteratively train the model using resources and calculate the loss on the model for the test set;

[0100] ③ Sort the test set losses of all hyperparameter combinations in ascending order; the smaller the loss, the better the combination.

[0101] ④ Only keep the best one from T. Each combination will enter the next cycle;

[0102] (4) Return the current best combination and the corresponding model.

[0103] Hyperband's multi-framework architecture allows each frame to have a different initial sample size n and initial iteration resource r, improving its robustness. Furthermore, during network iterative training, a stopping criterion is typically set after several consecutive iterations if the model performance shows no improvement, in order to conserve iteration resources.

[0104] In a preferred embodiment of the present invention, the method further includes,

[0105] The optimized engine life prediction model was used to predict the RUL value of the dataset. The predicted RUL value was evaluated using RMSE and Score evaluation metrics, and the prediction effect was evaluated.

[0106] The above preferred embodiments are summarized to obtain the following sub-steps of step 3:

[0107] The specific steps of step 3 include:

[0108] Step 3.1: After preprocessing the sensor data, the high-dimensional vector X is divided into a trend sequence X by passing it through a moving average filter. t and the remaining sequence X s H is then obtained by passing it through two linear layers and a convolutional layer respectively. t H s The summation yields the final prediction result Y. Its expression is as follows:

[0109]

[0110]

[0111] in:

[0112]

[0113] X s [i] = X[i] - X t [i] (5)

[0114] In the formula, Y represents the predicted remaining engine life, and W... t W t2 W s W s2 K represents the weights of the linear layer. t ,K s For the convolutional layer kernel, b t1 ,b s1 ,b tc ,b sc ,b t2 ,b s2 This is the bias term. Ds is the DLinear hyperparameter DKernalSize. X is divided into trend sequences X by a moving average filter. t and the remaining sequence X s Then, H is obtained by passing it through two linear layers and a convolutional layer respectively. t H s The two are added together to obtain the final prediction result.

[0115] W s W t These are the weight parameters for the two linear layers, b s b t is the bias term, and Ds is the hyperparameter used to segment the original sequence (DecompsitionKernelSize).

[0116] Step 3.2: Determine the hyperparameters to be tuned, their ranges and step sizes, and the configuration for Hyperband hyperparameter tuning;

[0117] The range and step size of the hyperparameters are shown in Table 1:

[0118]

[0119]

[0120] Table 1

[0121] The configuration is shown in Table 2:

[0122] Maximum number of resources R 81 η 3

[0123] Table 2

[0124] Step 3.3: Calculate the outer loop variable Total resources B = (s max +1)R, each s corresponds to one outer loop round, s∈{s max ,s max `-1,...,0}` represents the number of hyperparameter combinations for the external circulation cycle in this round. The actual maximum number of resources in the outer circulation of this round is r = Rη -s Obtain n sets of hyperparameter combinations from the hyperparameter space and store them in a sequence T. Each round of the outer loop contains a nested inner loop, with each i corresponding to one inner loop, where i∈{0,1,2,...,s}. Calculate the number of remaining hyperparameter combinations in the inner loop of that round. Calculate the amount of resources r that need to be allocated in this round of the loop. i =rη i For each combination of hyperparameters in T, assign r i Computational resources (here, computational resources are measured in iterations) are used for model training and testing. The test loss is obtained, and the models with the smallest loss in time interval T are selected. The optimal set of hyperparameters is obtained by carrying over some parameters to the next iteration until all iterations are completed. The results of selecting the optimal hyperparameters are shown in Table 3 below:

[0125]

[0126] Table 3

[0127] Step 3.3: Use the obtained Dlinear-CNN model to make predictions on the dataset to obtain the RUL prediction results.

[0128] Step 3.4: Use the root mean square error (RMSE) and score function to evaluate the RUL prediction results of Step 3.3. The smaller the RMSE and score values, the better the prediction effect.

[0129] The formula for calculating RMSE is:

[0130]

[0131] The score function is:

[0132]

[0133] In the formula, R represents the actual observed value (or true value) of the i-th sample. i Let N represent the predicted value of the i-th sample, where N is the number of samples.

[0134] in addition Figure 2 The image shown is a comparison of predicted and actual values ​​in the FD001 dataset when this invention is applied. Figure 2The results show that the predictive effect of the present invention has met expectations.

[0135] This invention also proposes a device for predicting the lifespan of aero-engines based on a Dlinear-CNN model, comprising the following:

[0136] The data acquisition module is used to acquire the original dataset of aircraft engine degradation, and to preprocess the original dataset to obtain processed data. The processed data is then used as sample data to divide the dataset into training set and test set.

[0137] The model training module is used to train a pre-built Dlinear-CNN model based on the training set, and to use the trained Dlinear-CNN model as an engine life prediction model.

[0138] The model testing module inputs the test set into the engine life prediction model after it is segmented by a sliding window. During the testing process, the Hyperband algorithm is used to optimize the hyperparameters to obtain the optimal hyperparameter combination. Finally, the predicted RUL value is obtained. The predicted RUL value is evaluated using RMSE and Score evaluation metrics, and the prediction effect is evaluated. The engine life prediction model that meets the preset standard evaluation result is taken as the optimized engine life prediction model.

[0139] The model deployment module is used to predict the lifespan of a target aero-engine using an optimized engine lifespan prediction model.

[0140] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0141] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0142] Although the description of the invention has been quite detailed and particularly of several described embodiments, it is not intended to limit it to any of these details or embodiments or any particular embodiment, but should be considered as providing a broad possible interpretation of the claims by referring to the appended claims and taking into account the prior art, thereby effectively covering the intended scope of the invention. Furthermore, the invention has been described above with respect to embodiments foreseeable by the inventors in order to provide a useful description, and non-substantial modifications to the invention that have not yet been foreseen may still represent equivalent modifications.

[0143] The above description is merely a preferred embodiment of the present invention. The present invention is not limited to the above-described embodiments. Any embodiment that achieves the technical effects of the present invention using the same means should fall within the protection scope of the present invention. Within the protection scope of the present invention, various modifications and variations can be made to the technical solutions and / or implementation methods.

Claims

1. A method for predicting the lifespan of aero-engines based on a Dlinear-CNN model, characterized in that, Including the following: Step 1: Obtain the original dataset of aircraft engine degradation, and perform data preprocessing on the original dataset to obtain processed data. Divide the processed data into training set and test set as sample data. Step 2: Train the pre-built Dlinear-CNN model based on the training set, and use the trained Dlinear-CNN model as the engine life prediction model; Step 3: After the test set is segmented by sliding window, it is input into the engine life prediction model for testing. During the testing process, the Hyperband algorithm is used to optimize the hyperparameters to obtain the optimal hyperparameter combination. The engine life prediction model is updated with the optimal hyperparameter combination to obtain the optimized engine life prediction model. Step 4: Use the optimized engine life prediction model to predict the life of the target aero-engine. Specifically, step 2 includes the following sub-steps: Step 2.1: Construct the Dlinear model; Step 2.2: Perform convolution on the input data to extract features from the data, select features, filter information, and combine features to obtain the final feature vector; Step 2.3: Integrate the DCNN layer into the Dlinear model to capture local features and sequence patterns. Set hyperparameters such as kernel size, learning rate, and sliding window length to train the model and obtain the trained Dlinear-CNN model. Specifically, the sliding window segmentation in step 3 includes, The high-dimensional vector obtained after data preprocessing of the test set data. The trend sequence is divided by a moving average filter. and remaining sequence Then, after passing through two linear layers and a convolutional layer respectively, we obtain... , The summation yields the final prediction result. The expression for Y is as follows: ； ； in: ； ； In the formula, Y represents the predicted remaining engine life. For linear layer weights, For convolutional layer kernels, Here, Ds is the bias term, and Ds is the DKernalSize hyperparameter of DLinear. Representative sequence Middle Each element.

2. The aero-engine life prediction method based on the Dlinear-CNN model according to claim 1, characterized in that, Specifically, step 1 includes the following sub-steps: Step 1.1: Set RUL labels for the initial sensor data in the original dataset to obtain processed data, and use a piecewise linear function to label the engine degradation mode; Step 1.2: Filter the processed data and remove data that is obviously irrelevant to the engine to obtain the filtered data; Step 1.3: Analyze the correlation between each sensor in the filtered data, filter and optimize the sensor data, and remove redundant sensor data to obtain deredundant data. Step 1.4: Perform data normalization on the deduplicated data to obtain normalized data, so that the processed data conforms to a standard normal distribution, i.e., the mean is 0 and the standard deviation is 1. The data normalization is shown in equation (1). ； In the formula, Represents the standardized eigenvalues. and Features The mean and standard deviation; Step 1.5: Remove outliers from the normalized data. Then, divide the processed data into training and test sets. Outlier detection uses... The method, as shown in equation (2), involves measuring the distance of a data point from the mean and comparing it to the standard deviation. ； In the formula, It is a dataset The mean, It is a dataset The standard deviation.

3. The aero-engine life prediction method based on the Dlinear-CNN model according to claim 1, characterized in that, Specifically, the labeled dataset is expanded using data augmentation strategies to obtain a training dataset, including: The labeled dataset is augmented using image enhancement techniques to obtain a training dataset. These image enhancement techniques include rotation, cropping, changing the hue, saturation, and brightness of the image, and adding noise.

4. The method for predicting the lifespan of aero-engines based on a Dlinear-CNN model according to claim 1, characterized in that, Specifically, step 3 uses the Hyperband algorithm to optimize hyperparameters and obtain the optimal hyperparameter combination, which includes: Determine the hyperparameters to be tuned, their ranges and step sizes, and the configuration for Hyperband hyperparameter tuning; Calculate the number of outer loops ,in The maximum number of iteration resources that can be allocated to a single combination of hyperparameters. The downsampling rate represents the best evaluation obtained after each iteration. Hyperparameter combinations can proceed to the next iteration, with a total resource count B = Each 's' corresponds to one round of the outer loop. { }, take the number of combinations of hyperparameters for the external circulation of that round. The actual maximum number of resources in this external circulation cycle n sets of hyperparameters are obtained from the hyperparameter space and combined into a sequence T. Each outer loop contains a nested inner loop, with each i corresponding to one inner loop. For each inner loop iteration, calculate the number of hyperparameter combinations to be evaluated ({0,1,2,...,s}). And the amount of resources that need to be allocated in this round of internal circulation. Assign hyperparameter combinations in T respectively Computational resources are used for model training and testing, resulting in test loss. The models with the smallest loss in time T are selected. The parameters are carried over to the next iteration until all iterations are completed, resulting in the optimal set of hyperparameters.

5. The method for predicting the lifespan of aero-engines based on a Dlinear-CNN model according to claim 1, characterized in that, The method also includes, The optimized engine life prediction model was used to predict the RUL value of the dataset. The predicted RUL value was evaluated using RMSE and Score evaluation metrics, and the prediction effect was evaluated.

6. A device for predicting the lifespan of aero-engines based on a Dlinear-CNN model, characterized in that, The apparatus comprising the steps of the method according to any one of claims 1-5, wherein the apparatus includes the following: The data acquisition module is used to acquire the original dataset of aircraft engine degradation, and to preprocess the original dataset to obtain processed data. The processed data is then used as sample data to divide the dataset into training set and test set. The model training module is used to train a pre-built Dlinear-CNN model based on the training set, and to use the trained Dlinear-CNN model as an engine life prediction model. The model testing module inputs the test set into the engine life prediction model after it is segmented by a sliding window. During the testing process, the Hyperband algorithm is used to optimize the hyperparameters to obtain the optimal hyperparameter combination. Finally, the predicted RUL value is obtained. The predicted RUL value is evaluated using RMSE and Score evaluation metrics, and the prediction effect is evaluated. The engine life prediction model that meets the preset standard evaluation result is taken as the optimized engine life prediction model. The model deployment module is used to predict the lifespan of a target aero-engine using an optimized engine lifespan prediction model.

Images