Remaining service life prediction method of high-dispersity cluster characteristic aero-engine based on callback model

Abstract

The invention provides a method for predicting the remaining service life of a high-dispersity cluster characteristic aero-engine based on a callback model, and solves the problem that gas path parameter values of engines of the same model are dispersed and distributed within a certain range due to manufacturing process level, working conditions, maintenance and the like of the aero-engine in actual engineering; and the residual service life of the engine is difficult to predict. Proposing and explaining high-dispersity cluster characteristics of aero-engine gas path parameters; constructing a multi-window output long-short-term memory neural network containing a convolution kernel, and fully utilizing all data to capture possible important information; defining an engine callback concept, and realizing re-learning of important features; and proposing a callback model based on the callback engine, defining a wedge blade matching algorithm as an optimal window selection criterion of each test engine in the callback model, and realizing prediction of the remaining service life of the engine with the high-dispersity cluster characteristic. According to the method for predicting the residual service life of the aero-engine, the performance difference between individual engines is fully considered, perfect data information is captured, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention proposes a method for predicting the remaining service life of an aeroengine with high dispersion cluster characteristics based on a callback model, belonging to the field of intelligent aeroengine health management. Background Art

[0002] The prediction and health management technology of aeroengines plays a crucial role in evaluating and determining the working state and life state of engines, as well as the safety and reliability of the entire aircraft. Predicting the remaining service life based on the obtained aeroengine health state monitoring data is a core technology in health management.

[0003] When conducting centralized analysis on the outfield flight data of the same type of aeroengines, it is found that the numerical values of the same gas path parameters among engines show a dispersed distribution within a certain range, and the differences are significant, showing high dispersion cluster characteristics. The reasons for this difference may include: 1) tolerances generated during the production and assembly of engines; 2) differences in the working hours, working states, and tasks performed by engines on different aircraft; 3) different degrees of noise pollution in the sensor data of each engine. These factors result in a large difference in the service lives among engine populations. After serving for the same period of time, some engines can still work normally for a long time, while some are already close to the usage threshold. Therefore, it is of great significance to propose a new method for predicting the remaining service life of engines suitable for high dispersion cluster characteristics.

[0004] The current mainstream algorithms for predicting the remaining useful life of aero-engines can be roughly divided into three categories: traditional methods for predicting the remaining useful life based on historical information, methods for predicting the remaining useful life based on physical models, and methods for predicting the remaining useful life based on data-driven approaches. The data-driven method can avoid establishing complex mathematical models, which has attracted wide attention. The patent with the publication number CN118350284A discloses a method for predicting the remaining useful life of complex equipment based on spatio-temporal feature fusion. It uses a graph convolutional network to extract the spatial features of the physical assembly relationship and fault relationship superposition of equipment components, adopts a bidirectional long short-term memory network and a multi-head attention mechanism to extract the full-life cycle time series features of sensor data, and finally fuses the spatio-temporal features to generate a remaining useful life prediction model. However, this method does not pay attention to and analyze the individual differences of the fleet. The patent with the publication number CN118657049A discloses a method for predicting the remaining useful life based on a dual convolutional attention mechanism network, which adds a dual convolutional attention mechanism to the prediction of the remaining useful life to enhance the attention to the time points of sensor data that have a great impact on the prediction of the remaining useful life, and fully considers the influence of different data time points on the RUL prediction. However, this method only uses the features learned from the last window data of the test samples for regression or classification, ignoring other features of the entire data trajectory of the engine. At present, there is no remaining useful life prediction method that focuses on the high-dispersion cluster characteristics of the engine, and the method for predicting the remaining useful life of aero-engines with high-dispersion cluster characteristics based on the callback model proposed in the present invention can effectively solve this problem. Summary of the Invention

[0005] Aiming at the problems existing in the existing data-driven methods, the present invention proposes a method for predicting the remaining useful life of aero-engines with high-dispersion cluster characteristics based on the callback model, so as to realize the prediction of the remaining useful life of engines with high-dispersion cluster characteristics.

[0006] To achieve the above object, the concept and technical solution of the present invention are realized as follows:

[0007] The basic concept of the present invention is to propose and explain the high-dispersion cluster characteristics of aero-engine gas path parameters, and separately explain the reasons why the gas path parameters of the same type of engine simultaneously have high dispersion and clustering; construct a multi-window output long short-term memory neural network with convolutional kernels, perform regression or classification on all window features of the test samples respectively, and make full use of all data to capture important information that may be contained; define the concept of a callback engine, find the callback engine for each window, and realize the relearning of important features; propose a callback model based on the callback engine, define the wedge-leaf matching algorithm as the best window selection criterion for each test engine in the callback model, and realize the prediction of the remaining useful life of engines with high-dispersion cluster characteristics. The model takes into account the performance differences between engine individuals, captures complete data information, and improves the prediction accuracy.

[0008] Based on the above basic concept, the technical solution proposed by the present invention is a method for predicting the remaining useful life of aero-engines with high-dispersion cluster characteristics based on a callback model, including the following steps:

[0009] Step 1: Propose and explain the high-dispersion cluster characteristics of aero-engine gas path parameters;

[0010] Step 2: Construct a multi-window output long short-term memory neural network with convolutional kernels;

[0011] Step 3: Define the concept of a callback engine to realize the relearning of important features;

[0012] Step 4: Propose a callback model based on the callback engine, and define the wedge-leaf matching algorithm as the best window selection criterion for each test engine in the callback model;

[0013] Step 5: Based on the callback model, accurately predict the remaining useful life of aero-engines with high-dispersion cluster characteristics.

[0014] Further, in the process of explaining the high-dispersion cluster characteristics of aero-engine gas path parameters in Step 1, it includes 3 operating condition parameters and 14 sensor parameters, specifically: the operating condition parameters are altitude alt, flight Mach number Mach, and throttle valve angle TRA; the sensor parameters are total temperature at the outlet of the low-pressure compressor T24, total temperature at the outlet of the high-pressure compressor T30, total temperature at the outlet of the low-pressure turbine T50, total pressure at the outlet of the high-pressure compressor P30, physical speed of the fan Nf, core physical speed Nc, static pressure at the outlet of the high-pressure compressor Ps30, ratio of fuel flow to Ps30 phi, corrected fan speed NRf, corrected core speed NRc, bypass ratio BPR, flow enthalpy htBleed, coolant flow of the high-pressure turbine W31, and coolant flow of the low-pressure turbine W32.

[0015] The coefficient of variation, like the standard deviation and variance, can reflect the degree of data dispersion. It is defined as the standard deviation σ and the mean value Ratio of:

[0016]

[0017] The coefficient of variation is a dimensionless index that can be used to compare two sets of data with different dimensions or different means. The coefficient of variation can be used to measure the dispersion of an aero-engine caused by factors such as manufacturing processes.

[0018] As the engine ages, its performance gradually degrades, and its state will be significantly different from the rated operating state at the time of factory shipment. At this time, the parameters of each gas path component measured by the sensor will change over time. The premise of gas path analysis and performance trend tracking is to obtain the health status of the engine from these changing parameters. In the research on predicting the remaining service life of an engine, the relative data of the parameter changes during the engine operation is more suitable for evaluating the dispersion degree of its performance. It is defined as:

[0019]

[0020] Where Δx i is the difference between the value of a certain parameter of the engine at the end of its life and the value at the beginning of its service, is the average value of all Δx i .

[0021] Furthermore, in step 2, a multi-window output long short-term memory neural network with a convolutional kernel is constructed, and regression or classification is performed on all window features to form a multi-input - multi-output model. This model makes full use of all the data of the test samples to capture important information.

[0022] Furthermore, in step 3, a callback engine is defined. Taking a test engine as an example, the engine trajectory is decomposed into multiple windows, which respectively correspond to the prediction results of the neural network, and the results of each window are used as features. Using these features, samples with similar features are searched for in the training samples, and these samples are the callback engines. Each window corresponds to a callback engine. The concept of callback is similar to the callback function in a talk show, forming a closed-loop topic logic, which is convenient for re-learning important features and enhancing the learning effect.

[0023] Furthermore, in step 4, a callback model is proposed based on the callback engine. All windows and their callback engines are used as the input of the model, and the output of the model is used as the prediction result of the remaining service life. The wedge leaf matching algorithm is defined as the best window selection criterion for each test engine in the callback model.

[0024] The cuneate leaf matching algorithm first adopts the naive Bayesian theory and regards all features as independent. The reason for naming it after the cuneate leaf is, on the one hand, the special shape of the cuneate leaf with whorled branches and leaves. The whole plant represents all the features contained in the windows of a test engine. Each node of the stem represents a window and its callback engine. The whorled branches at the stem nodes can represent the time series data of multiple parameters. The leaves of the whole plant contain the same judgment criterion. On the other hand, the cuneate leaf algorithm adopts the naive Bayesian theory, the fern classifier adopts the semi-naive Bayesian theory, and the decision tree adopts the Bayesian theory, with the complexity increasing gradually. This also just fits in with the evolutionary processes of the cuneate leaf, fern, and tree.

[0025] The cuneate leaf matching algorithm also includes the variance matching algorithm and the normalized correlation matching algorithm, which are respectively defined as:

[0026] Variance matching algorithm: E(p′ 2 ) - E 2 (p′) > a * E(p 2 ) - E 2 (p)

[0027] In the formula, E is the expected value, p′ represents the window of the training engine, p represents the window of the test engine, and a is the coefficient.

[0028] Normalized correlation matching algorithm:

[0029] In the formula, S is the window of the test engine; T is the window of the training engine. The sizes of both windows are M×N. M represents the length of the time series, and N represents the type of parameter. s mn and t mn respectively represent the elements in S and T. and respectively represent the average values of S and T. The result is in the range of [-1, 1]. The closer R is to 1, the greater the correlation.

[0030] Furthermore, in step 5, the remaining service life of the aeroengine with high dispersion cluster characteristics is accurately predicted based on the callback model. The specific steps are as follows:

[0031] Step 5.1: Use the multi-window output long short-term memory neural network with convolutional kernels to perform regression or classification on all windows of the test samples respectively, and obtain multiple prediction results;

[0032] Step 5.2: Find the callback engine for each window, and use all the window data, including the prediction results, together with the callback engine as the input of the callback model;

[0033] Step 5.3: Use the proposed matching algorithm to select the optimal window for each test engine, and the output of the model is used as the final prediction result.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] In the present invention, aiming at the problem that the flight data of engines of the same type are scattered within a certain range, which brings difficulties to data analysis, a method for predicting the remaining service life of an aeroengine with high-dispersion cluster characteristics based on a callback model is proposed. The high-dispersion cluster characteristics of the aeroengine are proposed and explained, and the influence of the dispersion degree of the gas path parameters of engines of the same model on the prediction of the remaining service life is emphasized; the concept of a callback engine is defined to help relearn important features and improve learning efficiency; the wedge-leaf matching algorithm is proposed, and the special shape of the wedge-leaf is used to vividly express the time series data contained in the engine to achieve accurate matching; the callback model based on the callback engine realizes the interaction between test samples and training samples, improving the accuracy of predicting the remaining service life. Description of the Drawings

[0036] Figure 1 is a flowchart of the method for predicting the remaining service life of an aeroengine with high-dispersion cluster characteristics based on a callback model provided by the present invention;

[0037] Figure 2 is a schematic diagram of a long short-term memory neural network with a convolutional kernel provided by the present invention;

[0038] Figure 3 is a schematic diagram of the way of capturing time series data by a sliding window provided by the present invention;

[0039] Figure 4 is a schematic diagram of predicting the remaining service life of a long short-term memory neural network with a convolutional kernel provided by the present invention;

[0040] Figure 5 is a schematic diagram of the result of a certain test engine based on a callback model provided by the present invention;

[0041] Figure 6 is a schematic diagram of predicting the remaining service life based on a callback model provided by the present invention. Detailed Embodiments

[0042] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0043] To verify the effectiveness of the proposed method, this example uses the publicly available C-MAPSS simulation dataset of NASA for verification. In this example, the FD001 dataset is used for model training and testing. Its training set contains data of 100 engines, and the test set contains data of 100 engines.

[0044] Please refer to Figure 1 , which shows the flowchart of a remaining useful life prediction method for aero-engines with high dispersion cluster characteristics based on a callback model provided by the present invention, specifically including the following steps:

[0045] Step 1: The process of explaining the high dispersion cluster characteristics of aero-engine gas path parameters, including 3 operating parameters and 14 sensor parameters, specifically: the operating parameters are altitude alt, flight Mach number Mach, and throttle valve angle TRA; the sensor parameters are total temperature at the outlet of the low-pressure compressor T24, total temperature at the outlet of the high-pressure compressor T30, total temperature at the outlet of the low-pressure turbine T50, total pressure at the outlet of the high-pressure compressor P30, physical speed of the fan Nf, physical speed of the core Nc, static pressure at the outlet of the high-pressure compressor Ps30, ratio of fuel flow to Ps30 phi, corrected fan speed NRf, corrected core speed NRc, bypass ratio BPR, flow enthalpy htBleed, coolant flow of the high-pressure turbine W31, and coolant flow of the low-pressure turbine W32.

[0046] Like the standard deviation and variance, the coefficient of variation can reflect the degree of data dispersion. It is defined as the ratio of the standard deviation σ to the mean value :

[0047]

[0048] The coefficient of variation is a dimensionless quantity and can be used to compare two sets of data with different dimensions or data with different mean values. The coefficient of variation is used to measure the dispersion of aero-engines caused by manufacturing processes, etc. The coefficients of variation of the 14 sensor parameters are shown in Table 1.

[0049] Table 1 Coefficients of variation of sensor parameters

[0050]

[0051] As the engine's performance gradually degrades with service time, its operating state will be significantly different from the rated state at the time of factory shipment. At this time, the parameters of each gas path component measured by the sensor will change with time. To conduct gas path analysis and performance trend tracking, the key lies in extracting the engine's health status from these changing parameters. In the research of engine health management, the relative change of data can more effectively evaluate the degree of dispersion of engine performance, which is defined as:

[0052]

[0053] where Δx i is the difference between the value of a certain engine parameter at the end of its service life and its value at the beginning of its service, and i is the average value of all Δx.

[0054] The c v,engine of the engine parameters is shown in Table 2. It can be seen that the maximum value of the c v,engine of these parameters is 1.33 and the minimum value is 0.26, indicating a relatively large degree of dispersion.

[0055] Table 2 c v，engine

[0056]

[0057] The main basis for gas path analysis and performance trend is the change of sensor parameters with the increase of service time, and the focus of the research is also on these change amounts. The degree of dispersion of the initial values of the parameters and the degree of dispersion of the subsequent time series data are much larger than the change amount Δx; for the parameters themselves, it can be seen from the coefficient of variation that their numerical distributions are relatively concentrated. Therefore, the engine sensor data exhibits the characteristics of a highly dispersed cluster.

[0058] Step 2: Construct a multi-window output long short-term memory neural network with convolutional kernels. Please refer to Figure 2 , the long short-term memory neural network with convolutional kernels extracts the features of data in the vicinity of space and time by using a convolutional structure in each unit. Set the number of learning units of the neural network to 100 and the learning rate to 0.001. Use the Adam optimizer and L1 regularization. The sliding window size is 15×30 and the step size is 1. Please refer to Figure 3 , the neural network uses a sliding window to extract two-dimensional samples, and the window moves along the time axis. The data of multiple time periods of the training samples are stacked for learning. The multi-window output long short-term memory neural network with convolutional kernels performs regression or classification on the features of all windows of the test samples respectively. The prediction results of all windows will be used as part of the input of the callback model, and the optimal window obtained under the proposed matching algorithm will be used as the final output result.

[0059] Step 3: Define and find the callback engine. The long short-term memory neural network with convolutional kernels learns the features of the training samples and predicts the remaining service life of the engines in the test set. The time series data of each engine is decomposed into multiple windows, and multiple remaining service life prediction values are obtained by predicting each window with the trained model. Use the predicted service life as the basis to find the engine with the same or closest service life in the training samples, and the found engine is the callback engine.

[0060] Step 4: Based on the callback engine, a callback model is proposed. All windows and their callback engines are used as inputs, and the model output is the predicted result of the remaining useful life. The wedge leaf matching algorithm is defined as the best window selection criterion for each test engine in the callback model.

[0061] The wedge leaf matching algorithm first adopts the Naive Bayes theory, considering all features as independent. Its name comes from the whorled form of the branches and leaves of the wedge leaf. The whole plant represents all the window features of a test engine. Each section of the stem corresponds to a window and its callback engine, while the branches represent the time series data of multiple parameters. The leaves of the whole plant contain the same judgment criteria.

[0062] The wedge leaf matching algorithm also includes the variance matching algorithm and the normalized correlation matching algorithm, which are defined as follows:

[0063] Variance matching algorithm: E(p′ 2 ) - E 2 (p′) > 0.7 * E(p 2 ) - E 2 (p)

[0064] In the formula, E is the expected value, p′ represents the training engine window, and p represents the test engine window. If the variance of the training engine window is greater than 70% of the variance of the test engine window, it passes.

[0065] Normalized correlation matching algorithm:

[0066] In the formula, S is the test engine window; T is the training engine window. The sizes of both windows are M×N. M represents the length of the time series, and N represents the type of parameter. s mn and t mn represent the elements in S and T respectively. and represent the averages of S and T respectively. The result is in the range of [-1, 1]. The closer R is to 1, the greater the correlation.

[0067] Step 5: Based on the callback model, an accurate prediction of the remaining useful life of aero-engines with high-dispersion cluster characteristics is realized. The evaluation indexes of this example are the root mean square error (RMSE) and the score function (Score), and the calculation formulas are as follows:

[0068]

[0069] In the formula, n represents the number of samples, Y i pre represents the predicted value of the i-th sample, and Y i represents the true value of the i-th sample.

[0070] The specific steps for predicting the remaining useful life of an aeroengine with high dispersion cluster characteristics based on the callback model are as follows:

[0071] Step 5.1: Use a multi-window output long short-term memory neural network with a convolutional kernel to perform regression or classification on all windows of the test samples respectively, and obtain multiple prediction results;

[0072] Step 5.2: Use the remaining useful life prediction results of all windows obtained in Step 5.1 as the basis for finding callback engines. Find callback engines for each window, and use all window data, including prediction results, and callback engines as the input of the callback model;

[0073] Step 5.3: Use the proposed matching algorithm to select the optimal window for each test engine, and the output of the model is used as the final prediction result.

[0074] Please refer to Figure 4 , which shows the prediction results of the remaining useful life of the test set engines using a long short-term memory neural network with a convolutional kernel. The RMSE of the prediction results is 12.76, and the Score is 289.

[0075] Please refer to Figure 5 , which shows the results of a certain test engine based on the callback model. If only the last window data of the engine trajectory is predicted, the difference between the predicted value and the actual value is 36, while the difference between the prediction result based on the callback model and the actual value is 2, greatly reducing the error.

[0076] Please refer to Figure 6 , which shows the prediction results of the remaining useful life of the test set engines using the callback model. The RMSE of the prediction results is 8.98, and the Score is 107. Compared with the model without callback, the RMSE is reduced by 22.49%, and the Score is reduced by 62.98%.

[0077] To prove the superiority of the present invention in predicting the remaining useful life of engines, the FD001 dataset of the CMAPSS dataset is used as the experimental object, and it is compared with existing common deep learning methods such as residual neural network (ResNet), long short-term memory neural network (LSTM), bidirectional long short-term memory neural network (Bi-LSTM), convolutional neural network (CNN), gated recurrent unit (GRU), and deep convolutional network - feedforward neural network (DCNN-FNN). It is further illustrated that the present model can effectively predict the remaining useful life and improve the prediction accuracy. The average index results of the comparison are shown in Table 3.

[0078] Table 3 Method Comparison

[0079] Method RMSE Score ResNet 16.62 376 LSTM 16.14 338 Bi-LSTM 14.59 296 CNN 18.45 1290 GRU 19.64 838 DCNN-FNN 12.61 273 The present invention 8.98 107

[0080] The present invention is not limited to the above embodiments. Based on the technical solutions disclosed in the present invention, those skilled in the art can make some simple modifications, equivalent changes and modifications to some of the technical features without creative labor, and all of them fall within the scope of the technical solutions of the present invention.

Claims

1. A method for predicting the remaining useful life of an aircraft engine with high-dispersion cluster characteristics based on a callback model, characterized in that: The specific steps include: Step 1: Propose and explain the high-dispersion cluster characteristics of aircraft engine gas path parameters; Step 2: Build a multi-window output long short-term memory neural network with convolution kernels; Step 3: Define the concept of callback engine to achieve re-learning of important features; Step 4: Propose a callback model based on the callback engine and define the wedge leaf matching algorithm as the best window selection criterion for each test engine in the callback model; Step 5: Based on the callback model, accurate prediction of the remaining useful life of aircraft engines with high-dispersion cluster characteristics is achieved.

2. The method for predicting the remaining useful life of an aircraft engine with high dispersion cluster characteristics based on a callback model according to claim 1 is characterized in that: The step 1 proposes and explains the high dispersion cluster characteristics of the air path parameters of the aircraft engine, and explains why the distribution of the air path parameters of the same type of engine has both high dispersion characteristics and cluster characteristics. In statistics, a statistic that characterizes the degree of dispersion of a probability distribution is the coefficient of dispersion, which is defined as the standard deviation σ and the mean value. The ratio: The dispersion coefficient is used to measure the clustering of gas path parameters of engines of the same model and the level of engine manufacturing technology. Considering that the change of each gas path component parameter over time is more important than the absolute value when evaluating the engine health, the discreteness of the parameter can be evaluated by analyzing the relative change data of the parameter during the engine operation: In the formula, Δx i It is the difference between the value of a parameter of the engine at the end of its service life and the value when it is just put into service. For all Δx i The average value of .

3. The method for predicting the remaining useful life of an aircraft engine with high dispersion cluster characteristics based on a callback model according to claim 1 is characterized in that: The step 2 constructs a multi-window output long short-term memory neural network with convolution kernels. The network regresses or classifies all window features of the test sample respectively, and the prediction results of all windows will be used as part of the input of the callback model.

4. The method for predicting the remaining useful life of an aircraft engine with high dispersion cluster characteristics based on a callback model according to claim 1 is characterized in that: Step 3 defines the concept of callback engine to achieve re-learning of important features. The engine trajectory is decomposed into multiple windows, and each window corresponds to the prediction results of the neural network. The result of each window is used as the feature of the window. Using this feature, the sample with the feature or the most similar feature is found in the training sample, and the found sample is the callback engine.

5. The method for predicting the remaining useful life of an aircraft engine with high dispersion cluster characteristics based on a callback model according to claim 1 is characterized in that: The step 4 proposes a callback model based on the callback engine, and defines a wedge leaf matching algorithm as the optimal window selection criterion for each test engine in the callback model. The algorithm is inspired by the wedge leaf shape, combined with the engine data characteristics, and adopts the naive Bayesian theory to achieve simple and fast preliminary matching.

6. The method for predicting the remaining useful life of an aircraft engine with high dispersion cluster characteristics based on a callback model according to claim 1 is characterized in that: The step 5 achieves accurate prediction of the remaining service life of aircraft engines with high-dispersion cluster characteristics based on the callback model. All windows of the test sample are regressed or classified respectively using a multi-window output long short-term memory neural network with a convolution kernel, and a callback engine is found for each window. All window data, including the prediction results, and the callback engine are used as inputs of the callback model, and the proposed matching algorithm is used to select the optimal window for each test engine as the final prediction result, so as to achieve the remaining service life prediction of the engine with high-dispersion cluster characteristics.

Images