Engine Performance Prediction Method Using Sample Adaptive Weighting

Abstract

The present invention relates to the technical field of engine condition monitoring, and particularly relates to an engine performance prediction method using sample adaptive weighting, which comprises the following steps: establishing a first data set and a second data set, wherein the first data set is the already-operated data set of the current engine, and the second data set includes the operation data sets of N other engines of the same model as the current engine; clustering the first data set and the second data set according to the flight envelope distribution to obtain M clustering centers; establishing sub-models such that each of the M clustering centers includes N sub-models; initializing a prediction weight vector W for each of the clustering centers, and performing weighted averaging on the N sub-models included therein to construct M linear weighted prediction models, which are used to predict engine performance; and optimizing the prediction weight vector W of the clustering centers. This method gives full play to data diversity and meets the requirements for accurate performance prediction of aero-engines.

Description

Technical Field

[0001] The present invention relates to the technical field of engine condition monitoring, and particularly to an engine performance prediction method using sample adaptive weighting. Background Art

[0002] The working environment of military aero-engines is extremely harsh. Environments such as high temperature, high pressure, and high speed cause the performance of each component to decline rapidly and the overall performance to degrade rapidly, which makes the life of military aero-engines relatively short. To ensure flight reliability, the performance of each engine needs to be monitored and predicted for timely maintenance, and it also helps to formulate a reasonable condition-based maintenance plan, thereby ensuring flight reliability and safety.

[0003] Engine performance monitoring and prediction usually complete mathematical modeling through complex operations using equations such as aerodynamics, rotor dynamics, and thermodynamics. However, the engine has a complex structure and numerous parts, and factors such as manufacturing and assembly errors are difficult to measure and estimate. The established performance model is an average model and does not contain any information about the individual differences of the engine. In addition, the engine working environment brings problems such as component performance degradation, and the performance degradation mechanism is extremely complex, which makes it very difficult to predict the engine. Therefore, in order to obtain a high-precision prediction model, characteristics such as engine individual differences and performance degradation need to be continuously corrected through the measured parameters of the engine.

[0004] Modeling methods based on data correction require rich and diverse data under a large number of working conditions to ensure the accuracy of the model. The commonly used engine adaptive models often require a certain amount of historical data. However, due to reasons such as the short life of military aero-engines, limited flight conditions, and low acquisition frequency of on-board storage units, when the engine running time is limited, the sample size is insufficient, and the flight conditions and states are not rich and diverse enough, it is difficult for its own samples to quickly and effectively correct the model.

[0005] Existing engine performance prediction methods lack the full utilization of diverse data from other engines and are also difficult to reflect the engine performance degradation caused by use, so it is difficult to meet the requirements of accurate performance prediction for modern military aero-engines. Summary of the Invention

[0006] Technical problem to be solved

[0007] In view of the above-mentioned drawbacks of the prior art, the present invention provides an engine performance prediction method using sample adaptive weighting, which can give full play to data diversity and meet the requirements of accurate performance prediction for aero-engines.

[0008] Technical solution

[0009] To achieve the above object, the present invention is realized by the following technical solutions:

[0010] The present invention provides an engine performance prediction method using sample adaptive weighting, including the following steps:

[0011] S1. Establish a first data set and a second data set. The first data set is the already-operated data set of the current engine, and the second data set includes the operation data sets of N other engines of the same model as the current engine;

[0012] S2. Cluster the first data set and the second data set according to the flight envelope distribution to obtain M clustering centers;

[0013] S3. N other engines respectively establish sub-models based on different clustering centers, so that each of the M clustering centers includes N sub-models;

[0014] S4. Initialize a prediction weight vector W for each clustering center, perform weighted averaging on the N sub-models it includes, and construct M linear weighted prediction models, which are used to predict engine performance;

[0015] S5. Initialize an empty data pool for each clustering center, put the current operation data of the current engine into the empty data pool of a clustering center, and optimize the prediction weight vector W of this clustering center. The one clustering center refers to the clustering center closest to the current operation data.

[0016] Further, step S2 specifically includes:

[0017] S2.1. Index all data samples in the first data set and the second data set by flight altitude and Mach number, randomly select k data samples as the initial cluster centers, and establish k clusters;

[0018] S2.2. For each remaining data sample, according to its distance from each initial cluster center, assign this sample to the closest cluster;

[0019] S2.3. Re-update the average value of each cluster and use it as the new center point of the cluster

[0020] S2.4. Continuously repeat S2.2 to S2.3 until no sample changes the cluster it belongs to or the cluster center point no longer changes. Each cluster is a clustering.

[0021] Further, the sub-model of each clustering center in step S3 is:

[0022]

[0023] Wherein, is the input vector, is the output vector, is the model function, and the subscript represents the serial number of the engine it belongs to, and the subscript represents the serial number of the cluster center it belongs to; M is the number of cluster centers; N is the number of other engines of the same model.

[0024] Furthermore, the linear weighted prediction model is:

[0025]

[0026] where is the performance prediction model function of the th cluster center; is the th element in the vector coefficient W, is the model function, and the subscript represents the serial number of the engine it belongs to, and the subscript represents the serial number of the cluster center it belongs to; M is the number of cluster centers; N is the number of other engines of the same model.

[0027] Furthermore, optimizing the prediction weight vector W of the cluster center specifically includes:

[0028] Calculate the output of the corresponding weighted prediction model by adding data to each empty data pool, establish a Levenberg-Marquardt optimizer, and use the error between the output calculated from the data pool and the actual output of the corresponding weighted prediction model as the optimization goal.

[0029] Furthermore, the Levenberg-Marquardt optimizer is:

[0030]

[0031] where represents the error vector between the output of the weighted prediction model and the output calculated from the data pool; W is the prediction weight vector; m is the interval time for parameter prediction.

[0032] Furthermore, it also includes step S6:

[0033] Repeat the process of S4 to S5 every m moments to adaptively optimize the prediction weight vector W in a rolling manner.

[0034] Furthermore, it also includes step S7: Plot the prediction performance curve to obtain the changing trend of the engine performance.

[0035] Furthermore, step S2 uses the K-means clustering method.

[0036] Beneficial effects

[0037] The present invention makes full use of information such as usage time and component performance degradation in the existing engine historical data for modeling, and at the same time takes into account information such as different flight conditions of multiple engines. Through clustering and scheduling, the data diversity is fully utilized. When the engine's own data is limited, the data-driven weighted prediction method is continuously optimized and iterated to gradually improve the prediction accuracy during the engine's use process. Description of the Drawings

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0039] Figure 1 Schematic diagram of the steps of the engine performance prediction method using sample adaptive weighting provided by an embodiment of the present invention;

[0040] Figure 2 Schematic flow chart of the engine performance prediction method using sample adaptive weighting provided by an embodiment of the present invention;

[0041] Figure 3 Another schematic flow chart of the engine performance prediction method using sample adaptive weighting provided by an embodiment of the present invention;

[0042] Figure 4 Schematic diagram of the initial data of the flight condition parameters of the training set provided by an embodiment of the present invention;

[0043] Figure 5 Schematic diagram of the result of clustering the flight condition parameters of the training set into 4 categories provided by an embodiment of the present invention;

[0044] Figure 6 Trend chart of the change of 4 clustering centers during the clustering process provided by an embodiment of the present invention;

[0045] Figure 7 Schematic diagram of the structure of the engine parameter prediction sub-model provided by an embodiment of the present invention;

[0046] Figure 8 Comparison chart between the engine exhaust temperature prediction model and the average model provided by an embodiment of the present invention. Detailed Embodiments

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part rather than all of the embodiments of the present invention. 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.

[0048] Referring to Figure 1 and Figure 2 , an embodiment of the present invention provides an engine performance prediction method using sample adaptive weighting, including the following steps:

[0049] S1. Establish a first data set and a second data set. The first data set is the already-operated data set of the current engine, and the second data set includes the operation data sets of N other engines of the same model as the current engine;

[0050] S2. Cluster the first data set and the second data set according to the flight envelope distribution to obtain M cluster centers;

[0051] S3. N other engines respectively establish sub-models based on different cluster centers, so that each of the M cluster centers includes N sub-models;

[0052] S4. Initialize a prediction weight vector W for each cluster center, perform weighted averaging on the N sub-models it includes, and construct M linear weighted prediction models, which are used to predict engine performance;

[0053] S5. Initialize an empty data pool for each cluster center, put the current operation data of the current engine into the empty data pool of one cluster center, and optimize the prediction weight vector W of this cluster center. The one cluster center refers to the cluster center closest to the current operation data.

[0054] During specific implementation, first, a data pool of existing operation data of other engines of the same model and the current engine data needs to be established, and data classification is achieved through clustering to reduce the model calculation dimension. Secondly, an average model is established using the data of other engines of the same model, and then the average model is transformed into a weighted model by optimizing the prediction deviation between the already-operated data of the current engine and the average model. Finally, the adaptive update of the weighting coefficient is realized through rolling optimization in the time domain, so as to continuously correct the engine prediction model.

[0055] In this embodiment, taking the exhaust gas temperature prediction of a certain type of turbofan engine as an example, the input variables of the prediction model are flight data such as flight altitude, Mach number, throttle lever angle, and operation time. In this embodiment, historical data of 5 engines are selected for test verification. Among them, 4 parts of data and the first 40% of the data of the engine to be predicted are used as the training data set of the prediction model, and the last 60% of the data of the engine to be predicted is used as the verification data set of the prediction model. Those skilled in the technical field of the present invention should understand that the type of the predicted engine and the prediction content are not limited, and this is only for illustration purposes.

[0056] In this embodiment, for step S2, using the K-means clustering algorithm, taking the historical data of engines of the same type and different parts as well as the flight envelope data of the currently operating engine as samples, as Figure 4 shown. After clustering in the figure, 4 clustering centers are obtained. Among them, the flight envelope refers to a closed geometric figure that uses parameters such as flight speed, altitude, overload, and ambient temperature as coordinates to represent the flight range of an aircraft and the usage limit conditions of the aircraft. Clustering refers to the process of dividing a set of physical or abstract objects into multiple classes composed of similar objects. The K-means clustering algorithm is an iterative clustering analysis algorithm. Its steps are to pre-divide the data into K groups, then randomly select K objects as the initial clustering centers, and then calculate the distance between each object and each seed clustering center, and assign each object to the clustering center closest to it. The clustering centers and the objects assigned to them represent a cluster.

[0057] In addition, as Figure 3 shown, step S2 specifically includes the following steps:

[0058] S2.1. Index all data samples in the first data set and the second data set by flight altitude and Mach number, randomly select k data samples as the initial center points of the clusters, and establish k clusters;

[0059] S2.2. For each remaining data sample, according to its distance from each of the initial center points of the clusters, assign this sample to the closest cluster;

[0060] S2.3. Re-update the average value of each cluster and use it as the new center point of the cluster

[0061] S2.4. Continuously repeat S2.2 to S2.3 until no sample changes the cluster it belongs to or the cluster center point no longer changes. Each cluster is a clustering. Among them, the cluster generated by clustering is a set of data objects, and these objects are similar to each other within the same cluster and different from the objects in other clusters.

[0062] For example, after 19 iterations in step S2.4, the clusters to which all samples belong remain unchanged, and the clustering is completed. The clustering result is as shown in Figure 5 , and the changes of the 4 clustering centers are as shown in Figure 6 .

[0063] In this embodiment, for step S3, N other engines respectively establish sub-models based on different ones of the clustering centers, so that each of the M clustering centers includes N sub-models. For example, the sub-models of each clustering center use the usage time, fuel flow rate, guide vane angle, and nozzle area as inputs and the temperatures of each cross-section of the engine as outputs to establish a neural network model. The sub-model of each clustering center can be:

[0064]

[0065] where, is the input vector, is the output vector, is the model function, and the subscript represents the serial number of the engine to which it belongs, and the subscript represents the serial number of the clustering center to which it belongs; M is the number of clustering centers; N is the number of other engines of the same model.

[0066] In this embodiment, the linear weighted prediction model is:

[0067]

[0068] where, is the performance prediction model function of the th clustering center; is the th element in the vector coefficient W; M is the number of clustering centers; N is the number of other engines of the same model.

[0069] In this embodiment, optimizing the prediction weight vector W of the clustering center specifically includes: calculating the output of the corresponding weighted prediction model by adding data to each empty data pool, establishing a Levenberg-Marquardt optimizer, and using the error between the output calculated from the data pool and the actual output of the corresponding weighted prediction model as the optimization target. And using the error between the output of the data pool and the output of the prediction model as the optimization target can be expressed as:

[0070]

[0071] where, represents the error vector between the output of the weighted prediction model and the output calculated from the data pool; W is the prediction weight vector; m is the interval time of parameter prediction.

[0072] In this embodiment, the structural schematic diagram of the engine temperature prediction sub-model is as follows Figure 7 shown. After calculation by the LM optimizer, the weighted vector solution is , Figure 8 which is the test comparison graph between the optimized prediction model using this weight and the unoptimized average prediction model ( ). The average error of the optimized prediction model is 0.95%, and the error of the unoptimized average prediction model is 1.5%. The method of the present invention is significantly better than the prediction model without using the optimized weight. When constructing the prediction model, the present invention is not limited to a certain model or type of military aviation engine and has a certain generality. By making full use of the richness and diversity of the historical data of different copies of the same model of military engines, this embodiment can predict parameters such as temperature, pressure, speed, and thrust, providing a research basis for subsequent possible life prediction, condition-based maintenance and other technologies.

[0073] The advantages of the present invention are that it makes full use of information such as service time and component performance degradation in the existing engine historical data for modeling, and at the same time takes into account information such as different flight conditions of multiple engines. Through clustering and scheduling, the data diversity is fully utilized. When the engine's own data is limited, the data-driven weighted prediction method is continuously optimized and iterated to gradually improve the prediction accuracy during the engine's use.

[0074] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. And these modifications or replacements will not make the essence of the corresponding technical solutions deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. An engine performance prediction method using sample adaptive weighting, characterized in that, it includes the following steps: S1. Establish a first data set and a second data set. The first data set is the operating data set of the current engine, and the second data set includes the operating data sets of N other engines of the same model as the current engine; S2. Cluster the first data set and the second data set according to the flight envelope distribution to obtain M cluster centers; S3. N other engines respectively establish sub-models based on different cluster centers, so that each of the M cluster centers includes N sub-models; S4. Initialize a prediction weight vector W for each cluster center, perform weighted averaging on the N sub-models it includes, and construct M linear weighted prediction models, which are used to predict engine performance; S5. Initialize an empty data pool for each cluster center, put the current operating data of the current engine into the empty data pool of one cluster center, and optimize the prediction weight vector W of this cluster center. The one cluster center refers to the cluster center closest to the current operating data; S6: Every m moments, repeat the process of S4 to S5 to adaptively optimize the prediction weight vector W in a rolling manner.

2. The engine performance prediction method using sample adaptive weighting according to claim 1, characterized in that, step S2 specifically includes: S2.

1. Index all data samples in the first data set and the second data set by flight altitude and Mach number, randomly select k data samples as the initial cluster centers, and establish k clusters; S2.

2. For each remaining data sample, assign the sample to the cluster with the closest distance according to its distance from each initial cluster center; S2.

3. Re-update the average value of each cluster and use it as the new cluster center point; S2.

4. Continuously repeat S1.2 to S1.3 until no sample changes the cluster it belongs to or the cluster center point no longer changes. Each cluster is a cluster.

3. The engine performance prediction method using sample adaptive weighting according to claim 1, characterized in that, the sub-model of each cluster center in step S3 is: Among them, is the input vector, is the output vector, is the model function, and the subscript represents the serial number of the engine it belongs to, and the subscript represents the serial number of the cluster center it belongs to; M is the number of cluster centers; N is the number of other engines of the same model.

4. The engine performance prediction method using sample adaptive weighting according to claim 1, characterized in that, the linear weighted prediction model is: Among them, is the performance prediction model function of the -th clustering center; is the -th element in the vector coefficient W, is the model function, the subscript represents the serial number of the engine to which it belongs, and the subscript represents the serial number of the clustering center to which it belongs; M is the number of clustering centers; N is the number of other engines of the same model.

5. The engine performance prediction method using sample adaptive weighting according to claim 1, characterized in that, the optimization of the prediction weight vector W of this cluster center specifically includes: Calculate the output of the corresponding weighted prediction model by adding data to each empty data pool, establish a Levenberg-Marquardt optimizer, and use the error between the output calculated from the data pool and the actual output of the corresponding weighted prediction model as the optimization target.

6. The engine performance prediction method using sample adaptive weighting according to claim 5, characterized in that, the Levenberg-Marquardt optimizer is: wherein, represents the error vector between the output of the weighted prediction model and the output calculated by the data pool; W is the prediction weight vector; and m is the interval time for parameter prediction.

7. The engine performance prediction method using sample adaptive weighting according to claim 1, characterized in that, it further includes step S7: plotting a predicted performance energy curve to obtain the change trend of the engine performance energy.

8. The engine performance prediction method using sample adaptive weighting according to claim 1, characterized in that, step S2 uses the K-means clustering method.

Images