Training methods, diagnostic methods and devices for aircraft engine fault diagnosis models

By optimizing the features of aero-engine performance parameters and constructing a fault diagnosis model using the KNN algorithm, the problem of complex and inefficient aero-engine fault diagnosis in existing technologies has been solved, achieving efficient and accurate fault identification.

CN115994479BActive Publication Date: 2026-11-17XIAMEN UNIV +1
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202211284284.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-10-20
Publication Date
2026-11-17
Estimated Expiration
2042-10-20

AI Technical Summary

Technical Problem

Existing methods for diagnosing aero-engine faults are complex and inefficient, and cannot effectively identify faults in aero-engines under extreme environments.

Method used

By obtaining the performance parameter deviation values ​​of aero-engines, performing feature optimization, and then using the KNN algorithm to construct a fault diagnosis model, including preprocessing of the original data, feature dimensionality enhancement and filtering, combined with minority class oversampling and data normalization processing, fault diagnosis is achieved.

Benefits of technology

It improves the efficiency and accuracy of aircraft engine fault diagnosis, reduces reliance on physical structure, and enhances the accuracy and speed of fault identification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115994479B_ABST
    Figure CN115994479B_ABST
Patent Text Reader

Abstract

This invention provides a method, diagnostic method, and apparatus for training an aero-engine fault diagnosis model. It includes: acquiring deviation values ​​of aero-engine performance parameters; performing feature optimization on the deviation values ​​of the performance parameters to obtain optimized performance parameter deviation values; and using the optimized performance parameter deviation values ​​as input to construct an aero-engine fault diagnosis model based on the KNN algorithm. This invention, through its method, diagnostic method, and apparatus, optimizes the deviation values ​​of aero-engine performance parameters and uses the optimized deviation values ​​as input to construct an aero-engine fault diagnosis model based on the KNN algorithm. Finally, it diagnoses aero-engine faults using this model. This method no longer relies on the physical structure of the aero-engine, thus improving diagnostic efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of aviation technology, and in particular to a training method, diagnostic method and device for an aircraft engine fault diagnosis model. Background Technology

[0002] Aero engines are highly complex power devices, and aircraft flight safety and availability depend on the condition of their engines. During flight, aero engines operate continuously for extended periods in extremely harsh environments characterized by strong vibrations and high noise levels, inevitably leading to certain malfunctions. Such malfunctions can result in significant economic losses and personal injury. Therefore, researching aero engine fault diagnosis methods and conducting comprehensive fault diagnosis is of paramount engineering value in ensuring the normal and safe operation of aero engines.

[0003] However, current methods for diagnosing aero-engine faults still require consideration of the engine's physical structure, making the diagnostic process complex and inefficient. With the rapid development of computer science and intelligent information processing technology, if data-driven fault diagnosis methods could be adopted, the physical structure of the aero-engine would not need to be considered. Therefore, researching data-driven methods for aero-engine fault diagnosis is of great significance. Summary of the Invention

[0004] This invention provides a training method, diagnostic method, and apparatus for aero-engine fault diagnosis models, which addresses the shortcomings of existing technologies in diagnosing aero-engine faults due to insufficient efficiency, and achieves efficient diagnosis of aero-engine faults.

[0005] The present invention provides a training method for an aero-engine fault diagnosis model, comprising: obtaining deviation values ​​of performance parameters of the aero-engine, wherein the performance parameters include: exhaust gas temperature (EGT), high-pressure turbine speed N2, and fuel flow (FF);

[0006] The deviation values ​​of the performance parameters are subjected to feature optimization to obtain optimized performance parameter deviation values;

[0007] Using the optimized performance parameter deviation values ​​as input, an aero-engine fault diagnosis model is constructed based on the KNN algorithm.

[0008] According to the training method for a fault diagnosis model of an aero-engine provided by the present invention, the performance parameters further include: high-pressure compressor outlet temperature T3 and low-pressure compressor outlet temperature T25.

[0009] According to the present invention, a method for training an aero-engine fault diagnosis model includes obtaining the deviation values ​​of the aero-engine performance parameters.

[0010] Obtain test performance parameters of aero engines;

[0011] Baseline modeling is performed based on a multivariate linear method to obtain baseline performance parameters of aero-engines;

[0012] The deviation value of the performance parameter is obtained by subtracting the baseline performance parameter from the measured performance parameter.

[0013] According to the present invention, a training method for an aero-engine fault diagnosis model includes preprocessing the original performance parameters before baseline modeling, and using the preprocessed performance parameters as the basis for baseline modeling.

[0014] The pretreatment measures include:

[0015] The nearest neighbor imputation method was used to fill in all missing values ​​in the original performance parameter data to obtain complete performance parameter data;

[0016] Extract the performance parameter data of the aircraft cruise segment from the complete performance parameter data;

[0017] Data from a period after water washing was selected from the performance parameter data of the aircraft cruise segment for baseline modeling.

[0018] According to the present invention, a training method for an aero-engine fault diagnosis model includes feature optimization of the deviation values ​​of the performance parameters to obtain optimized performance parameter deviation values, which includes:

[0019] The deviation values ​​of the performance parameters are augmented with features to obtain the augmented deviation values ​​of the performance parameters.

[0020] The optimal performance parameter deviation values ​​are obtained by filtering the performance parameter deviation values ​​after dimensionality enhancement.

[0021] The optimal performance parameter deviation value is used as the basis for constructing the aero-engine fault diagnosis model.

[0022] The present invention also provides a method for diagnosing aircraft engine faults, comprising:

[0023] Acquire data on the aircraft engine faults to be diagnosed;

[0024] The fault data of the aircraft engine to be diagnosed is input into a pre-built aircraft engine fault diagnosis model to obtain the diagnosis results;

[0025] The diagnostic results include: what type of fault the aircraft engine fault data to be diagnosed belongs to;

[0026] The aircraft engine fault diagnosis model is constructed using any of the methods described above.

[0027] According to the aircraft engine fault diagnosis method provided by the present invention, the step of acquiring the aircraft engine fault data to be diagnosed includes:

[0028] Acquire raw fault data of aircraft engines;

[0029] The original fault data is augmented using a minority class oversampling method to obtain augmented fault data;

[0030] The expanded fault data is normalized to obtain normalized fault data.

[0031] The normalized fault data is used as the fault data of the aero-engine to be diagnosed.

[0032] The present invention also provides an aero-engine fault diagnosis model training device, comprising:

[0033] The acquisition unit is used to acquire the deviation values ​​of the performance parameters of the aero-engine, including: exhaust temperature EGT, high-pressure turbine speed N2, and fuel flow rate FF;

[0034] An optimization unit is used to perform feature optimization on the deviation values ​​of the performance parameters to obtain optimized performance parameter deviation values.

[0035] The construction unit is used to construct an aero-engine fault diagnosis model based on the KNN algorithm, using the optimized performance parameter deviation value as input.

[0036] The present invention also provides an aero-engine fault diagnosis model training device, comprising:

[0037] Acquisition unit, used to acquire fault data of the aircraft engine to be diagnosed;

[0038] The prediction unit is used to input the aircraft engine fault data to be diagnosed into a pre-built aircraft engine fault diagnosis model for prediction and to obtain the diagnosis result.

[0039] The diagnostic results include: which type of fault the aircraft engine fault data to be diagnosed belongs to; and the aircraft engine fault diagnosis model is constructed using the method described above.

[0040] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aircraft engine fault diagnosis model training method as described above or the aircraft engine fault diagnosis method as described above.

[0041] Beneficial effects:

[0042] The present invention provides an aero-engine fault diagnosis model training method, a diagnostic method and device. By optimizing the deviation values ​​of the performance parameters of the aero-engine, and using the optimized performance parameter deviation values ​​as input, an aero-engine fault diagnosis model is constructed based on the KNN algorithm. Finally, the aero-engine fault diagnosis model is used to diagnose aero-engine faults. This method no longer depends on the physical structure of the aero-engine, thereby improving the efficiency of diagnosis. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0044] Figure 1 The flowchart of the aero-engine fault diagnosis model training method provided by the present invention is shown below.

[0045] Figure 2 Let be the objective function of NCA under different regularization coefficients λ;

[0046] Figure 3 The feature weights are for different features;

[0047] Figure 4(a) shows a comparison of the accuracy of tree-based algorithms for different feature groups;

[0048] Figure 4(b) shows a comparison of the accuracy of different feature groups based on the KNN algorithm;

[0049] Figure 5 The image shows the optimization results of the KNN algorithm.

[0050] Figure 6(a) shows the confusion matrix of refined KNN using features selected based on correlation.

[0051] Figure 6(b) shows the ROC curves for features selected using correlation.

[0052] Figure 7 Diagnostic flowchart for the constructed aircraft engine fault diagnosis model;

[0053] Figure 8 This is a diagram illustrating the initial data extraction process for the cruise segment.

[0054] Figure 9 This is a schematic diagram of cruise point extraction;

[0055] Figure 10 This is a schematic diagram illustrating the extraction of water washing data from QAR baseline modeling data.

[0056] Figure 11 A flowchart of the aircraft engine fault diagnosis method provided by the present invention;

[0057] Figure 12 This diagram illustrates the impact of normalization on the model's convergence speed.

[0058] Figure 13 A complete flowchart for constructing the fault diagnosis model for this invention;

[0059] Figure 14 This is a schematic diagram of the structure of the aero-engine fault diagnosis model training device of the present invention;

[0060] Figure 15 A schematic diagram of the structure of an aircraft engine fault diagnosis device;

[0061] Figure 16 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0063] Figure 1 The flowchart of the aero-engine fault diagnosis model training method provided by the present invention is as follows: Figure 1 As shown, the method includes the following steps:

[0064] Step 101: Obtain the deviation values ​​of the performance parameters of the aero-engine, including: exhaust temperature EGT, high-pressure turbine speed N2, and fuel flow rate FF.

[0065] Specifically, the deviation values ​​of the performance parameters include: the deviation value of exhaust temperature △EGT, the deviation value of high-pressure turbine speed △N2, and the deviation value of fuel flow rate △FF. This invention uses the deviation values ​​of these performance parameters to assess the faults of aero-engines; hereinafter, the deviation values ​​of the performance parameters will be referred to simply as fault characteristics.

[0066] To improve the accuracy of fault diagnosis for aero-engines, this invention adds the deviation values ​​ΔT3 of the high-pressure compressor outlet temperature and ΔT25 of the low-pressure compressor outlet temperature as parameters for judging whether an aero-engine has a fault, in addition to the original performance parameters.

[0067] The following is a detailed explanation of how to obtain the deviation values ​​of the performance parameters of an aero-engine:

[0068] First, a baseline model was constructed using a multiple linear regression method, resulting in a multiple linear regression equation, as shown in formula (1). The independent variables for this baseline were selected as low-pressure turbine speed N1, total air temperature (TAT), Mach number M, and flight altitude ALT. The baseline was established with EGT, N2, FF, T3, and T25.

[0069]

[0070] The specific expressions for the five baselines in formula (1) are shown in formula (2), where a to e are the undetermined coefficients of the linear regression equation.

[0071]

[0072] Where a1~a5, b1~b5, c1~c5, d1~d5, and e1~e5 are the undetermined coefficients of the linear regression equation;

[0073] Secondly, based on the given baseline modeling data, the undetermined coefficients of the baseline equation in equation (2) are determined by the least squares method.

[0074] Finally, after obtaining the baseline equation, the deviation value of the performance parameters of the aero-engine can be obtained by subtracting the baseline performance parameters from the measured performance parameters of the aero-engine, as shown in equation (3).

[0075]

[0076] Step 102: Perform feature optimization on the deviation values ​​of the performance parameters to obtain optimized performance parameter deviation values.

[0077] Specifically, the five parameters △EGT, △N2, △FF, △T3, and △T25 contain fault information, but if directly input into a fault diagnosis model, the model's prediction performance is poor. Therefore, to further extract the hidden information, this invention first employs various dimensionality enhancement methods to increase the number of original fault features. However, while dimensionality enhancement amplifies the fault information, it also introduces a lot of redundant and ineffective information. Therefore, a dimensionality reduction method is then used to filter the dimensionality-enhanced fault features to obtain the optimal combination of fault features. This invention refers to the two processes of feature dimensionality enhancement and feature selection as feature optimization.

[0078] The following explains how to perform feature augmentation on fault characteristics:

[0079] To increase the number of fault features to achieve feature dimensionality enhancement, this invention can employ various methods to achieve feature dimensionality enhancement. For example, this invention uses exponential feature dimensionality enhancement, logarithmic feature dimensionality enhancement, combined feature dimensionality enhancement, and difference feature dimensionality enhancement to enhance the fault features △FF, △N2, △EGT, △T3, and △T25. The results of feature dimensionality enhancement are shown in Table 1, and there are a total of 36 fault features after dimensionality enhancement.

[0080] Table 1. Feature Dimensioning Results

[0081]

[0082] The following explains how to filter fault characteristics after dimensionality enhancement:

[0083] This invention can employ various screening methods to filter the fault features after dimensionality enhancement. For example, this invention uses nearest neighbor principal component analysis to select the optimal feature combination from 36 features.

[0084] The specific implementation steps of NCA are as follows:

[0085] Step 1: Let w r Define the distance function d for the distance weights of the features. w As in equation (4):

[0086]

[0087] Step 2: Let k be a similarity function, and define the reference sample of sample x as xi. j The probability is as shown in (5):

[0088]

[0089] Here, NCA assumes that x and x j When the distance is small, x j The sample with the higher probability is used as the reference sample for x. When xj When x is a reference sample, use x j The tag y j As a label for x.

[0090] Step 3: Remove samples (x) i ,y i After that, x is calculated using equation (6). i The reference sample is x j The probability of removing one:

[0091]

[0092] Step 4: Calculate the reference sample x j The tag and x i The probability of the same label is given by equation (7). Obviously, when p i The larger the value, the smaller the distance between samples, and the greater the probability that they are used as reference samples, meaning there is a greater probability that they belong to the same class.

[0093]

[0094] Step 5: Calculate the average of the "one-out probability" of all samples as shown in Equation (8), and construct the objective function by adding a regularization term composed of λ as shown in Equation (9). After constructing the objective function, the feature selection problem is transformed into a minimum value problem as shown in Equation (10).

[0095]

[0096]

[0097]

[0098] The 36 features are numbered according to the order listed in the table. Nearest neighbor principal component analysis is then used to filter the dimensionality-enhanced fault features, selecting the optimal feature combination (obtained from the 36 features). The objective function and the regularization coefficient λ = Lambda are then constructed. Figure 2 The graph shows the objective function of NCA under different regularization coefficients λ. Figure 3 For different feature weights, when the regularization coefficient λ is selected... * =8.0091*10 -5 At time (solid dots in the diagram), the objective function value is at its lowest, and the weights of each feature at this time are as follows: Figure 3 As shown.

[0099] Choose the regularization coefficient λ * =8.0091*10 -5 The weight values ​​of each feature at time are as follows Figure 3 As shown in Table 2, with a threshold of 0.001, the final fault characteristics retained are as follows.

[0100] Table 2 NCA Feature Preservation Results

[0101]

[0102] Step 103: Using the optimized performance parameter deviation value as input, construct an aero-engine fault diagnosis model based on the KNN algorithm.

[0103] To highlight the importance of feature optimization, this invention divides fault features into three groups, as shown in Table 3, and compares them with fault diagnosis algorithms based on KNN, Naive Bayes, and trees, as shown in Table 4. Fault diagnosis models for aero-engines are constructed respectively, and the accuracy of the models is predicted by the constructed models.

[0104] Table 3 Fault Feature Classification Table

[0105]

[0106] Table 4. Classification algorithms used to build fault models

[0107]

[0108] The results using the algorithm in Table 4 are shown in Tables 5-7. The refined KNN model achieves optimal diagnostic accuracy when using different feature sets. When using the original features ΔEGT, ΔFF, and ΔN², the model built based on refined KNN achieves a diagnostic accuracy of 98.3%; when using ΔEGT, ΔFF, ΔN², and augmented features, the refined KNN model achieves a diagnostic accuracy of 98.4%; and when using ΔEGT, ΔFF, ΔN², ΔT³, ΔT²⁵, and augmented features, the refined KNN model achieves a diagnostic accuracy of 98.6%.

[0109] Table 5 shows the diagnostic results using △EGT, △FF, and △N2.

[0110]

[0111]

[0112] Table 6 shows the feature diagnosis results after dimensionality enhancement and optimization using △EGT, △FF, and △N2.

[0113]

[0114] Table 7 shows the feature diagnosis results after dimensionality enhancement and optimization using △EGT, △FF, △N2, △T3, and △T25.

[0115]

[0116] To compare the impact of different feature groups on algorithm accuracy, accuracy comparison charts were drawn. Figure 4(a) shows the accuracy comparison chart of different feature groups based on the tree algorithm; Figure 4(b) shows the accuracy comparison chart of different feature groups based on the KNN algorithm. As shown in Figures 4(a) and 4(b), when only ΔEGT, ΔFF, and ΔN2 are used, the fault diagnosis accuracy is low under different algorithms. When dimensionality-enhancing features are added to ΔEGT, ΔFF, and ΔN2, the fault diagnosis accuracy is improved compared to using only ΔEGT, ΔFF, and ΔN2. The highest fault diagnosis accuracy can be obtained after using ΔEGT, ΔFF, ΔN2, ΔT3, ΔT25, and dimensionality-enhancing features. This demonstrates that the introduction of ΔT3, ΔT25, and feature dimensionality enhancement helps improve fault diagnosis accuracy.

[0117] This invention further optimizes the KNN algorithm to determine the optimal KNN algorithm parameters. The optimized parameter ranges are as follows: number of neighboring points: 1 to 6822; distance metric: City Block, Chebyshev, Cosine, Euclidean, Hamming, Jaccard. Figure 5 The image shows the optimization results of the KNN algorithm. Figure 5 As shown, the final number of neighboring points selected is 1, and the distance metric chosen is Euclidean distance.

[0118] Figure 6(a) shows the confusion matrix of the refined KNN using features selected based on correlation, and Figure 6(b) shows the ROC curve of the refined KNN using features selected based on correlation. The confusion matrix for fault diagnosis under the optimal KNN algorithm is shown in Figure 6(a), indicating that a significant number of Class 1 (no-fault) faults are misdiagnosed as faults. The algorithm's false alarm rate is 5.4%, and the false negative rate is 0%. For Class 2 (fuel pump fault), Class 3 (bird strike), and Class 4 (fan lubrication fault), the algorithm achieves a fault detection rate of 100%, meeting the technical requirements. The ROC and AUC of this algorithm are shown in Figure 6(b). The AUC of this algorithm is 0.97, indicating that the refined KNN performs well.

[0119] Figure 7 The diagnostic flowchart for the constructed aero-engine fault diagnosis model is as follows: Figure 7 As shown, the diagnostic process includes the following steps:

[0120] Step 1: Initialize the KNN algorithm and determine the nearest neighbor number K and the distance calculation method in the KNN algorithm;

[0121] Step 2: Then, calculate the distance between the test sample and each other sample based on the defined distance.

[0122] Step 3: Sort the calculated distances from smallest to largest, and select the top K sample points from them; perform statistical analysis on the categories of the K samples, and determine which category the test sample belongs to if the number of samples in that category is the largest.

[0123] The present invention provides an aero-engine fault diagnosis model training method, a diagnostic method and device. By optimizing the deviation values ​​of the performance parameters of the aero-engine, and using the optimized performance parameter deviation values ​​as input, an aero-engine fault diagnosis model is constructed based on the KNN algorithm. Finally, the aero-engine fault diagnosis model is used to diagnose aero-engine faults. This method no longer depends on the physical structure of the aero-engine, thereby improving the efficiency of diagnosis.

[0124] Furthermore, prior to baseline modeling, the original performance parameters are preprocessed, and the preprocessed performance parameters serve as the basis for baseline modeling.

[0125] The pretreatment measures include:

[0126] The nearest neighbor imputation method is used to fill in all missing values ​​in the original performance parameter data to obtain complete performance parameter data; the performance parameter data of the aircraft cruise segment is extracted from the complete performance parameter data; and data within a certain period after water washing is selected from the performance parameter data of the aircraft cruise segment for baseline modeling.

[0127] The following explains how to fill in the original performance parameter data.

[0128] Specifically, firstly, since the independent variables N1, TAT, M, and ALT in the multiple linear regression model are derived from engine QAR data during baseline modeling, and the sensor signals are periodically missing due to different sampling frequencies of different sensors, this invention employs a nearest neighbor imputation method to fill in all missing values ​​based on the similarity between missing values ​​and nearby values ​​in engine QAR data. This makes the relevant data of independent variables N1, TAT, M, and ALT more complete, thereby further improving the prediction accuracy of the fault model.

[0129] The following explains how to extract performance parameter data for the aircraft cruise segment from complete performance parameter data.

[0130] During takeoff, climb, and landing, the engine's parameters change significantly due to the constantly changing external environment and control variables, which makes data analysis difficult. Therefore, before baseline modeling, data from the aircraft's cruise phase is extracted and used as the basis for baseline modeling to further improve the predictive accuracy of the fault model.

[0131] The data extraction process for the aircraft cruise segment is as follows:

[0132] Step 1: Initially filter cruise segment data based on parameters such as Calculated Air Speed ​​(CAS), Mach number (M), and fuel flow rate (FF). During flight, some parameters, such as CAS, MACH, FF, N1, and N2, remain within a certain range. If the range of parameter changes exceeds this threshold, the aircraft is considered to be in a non-stable flight phase (such as the ground phase). Therefore, CAS, MACH, FF, N1, and N2 are used as discrimination parameters to initially extract data for the aircraft's cruise segment. The extraction rules are shown in Table 3. Initial filtering based on the parameter value ranges in Table 3 will yield the desired data. Figure 8 , Figure 8 This is a diagram illustrating the preliminary extraction process of cruise segment data; the extraction results are as follows. Figure 8 As shown.

[0133] Table 3 Preliminary Extraction of Aircraft Cruise Segment Rules

[0134]

[0135] Step 2: Set the ALT deviation to no more than 10 feet. The extracted cruise segment data is as follows: Figure 9 As shown, the span of the extracted cruise segment data is determined sequentially. The first cruise segment data with a span exceeding 500 cycles is selected, and data from one point is randomly selected as the cruise point data, as shown below. Figure 9 As shown by the solid dot.

[0136] The following explains how to filter data from the performance parameter data of an aircraft during its cruise phase, specifically data for a period of time after washing.

[0137] Since the baseline model obtains the engine's parameters under healthy conditions, comparing these parameters with existing parameters reveals fault characteristics, i.e., deviations in performance parameters. Therefore, this invention uses data from a period after water washing for baseline modeling. Figure 10 As shown, since the current QAR data spans from January 2015 to July 2018 and does not include data from when the engine first entered service, this invention filters data from a period after water washing for baseline modeling.

[0138] The QAR data used for baseline modeling follows these three principles:

[0139] ① Select QAR data within a certain period of time after water washing;

[0140] ② Use QAR data as early as possible to avoid engine degradation;

[0141] ③ Select enough data to ensure that accurate baseline values ​​can be obtained under different operating conditions.

[0142] This invention also provides a method for diagnosing aircraft engine faults. Figure 11 The flowchart of the aircraft engine fault diagnosis method provided by the present invention is as follows: Figure 11 As shown, the method includes the following steps:

[0143] Step 1101: Obtain the fault data of the aircraft engine to be diagnosed.

[0144] Before obtaining the fault data of the aircraft engine to be diagnosed, it is necessary to first preprocess the raw aircraft engine fault data to obtain the fault data of the aircraft engine to be diagnosed.

[0145] The following section provides a detailed explanation of how to preprocess raw aircraft engine fault data.

[0146] Obtain the original fault data of the aero-engine; augment the original fault data using a minority class oversampling method to obtain augmented fault data; normalize the augmented fault data to obtain normalized fault data; use the normalized fault data as the fault data of the aero-engine to be diagnosed.

[0147] Specifically, because the probability of fault samples occurring during actual flight is extremely low, the number of aircraft engine fault data samples to be diagnosed is less than the number of healthy samples. This imbalance between samples of different states affects the model's predictive bias. The model's predictions will tend to diagnose states with more samples, resulting in poor generalization and accuracy when the model is applied to new data. To eliminate the problem caused by data imbalance, this invention requires sample equalization of the aircraft engine fault data to be diagnosed before performing model prediction. The Synthetic Minority Over-sampling Technique (SMOTE) is used to expand the minority samples, thereby achieving sample equalization.

[0148] The specific methods for sample equalization are described below:

[0149] The basic idea of ​​the SMOTE method is to analyze and simulate minority class samples, and then add artificially simulated new samples to the dataset, thereby making the classes in the original data no longer severely imbalanced.

[0150] The steps to implement SMOTE are as follows:

[0151] Step 1: Use the nearest neighbor algorithm to calculate the K (K=10) nearest neighbors for each minority class sample;

[0152] Step 2: Select N (N=5) samples from the K nearest neighbors in Step 1 and perform random linear interpolation;

[0153] Step 3: Construct new samples based on the results of random linear interpolation in Step 2.

[0154] The following explains the methods for data normalization:

[0155] Because the different data have different dimensions, the numerical values ​​of each data type vary greatly. For example, the maximum value of altitude data can reach about 38,000 feet, while the maximum value of Mach number data is only about 0.7 M. The difference in their numerical dimensions and the huge gap in magnitude are detrimental to the convergence speed and accuracy of the model. Therefore, this invention performs data normalization processing on various data. Figure 12 This is a schematic diagram illustrating the impact of normalization on model convergence speed, as shown below. Figure 12 As shown, the normalized parameters have a faster convergence speed under the same conditions.

[0156] The normalization method used in this invention is Z-score normalization, and its calculation formula is shown in (11):

[0157]

[0158] In the formula X * X is the normalized value of a certain feature; max and X min These are the maximum and minimum values ​​of this feature.

[0159] Step 1102: Input the aircraft engine fault data to be diagnosed into the pre-built aircraft engine fault diagnosis model to obtain the diagnosis result; the diagnosis result includes: which type of fault the aircraft engine fault data to be diagnosed belongs to; the aircraft engine fault diagnosis model is constructed using the method described above.

[0160] Example:

[0161] This embodiment will provide an overall introduction to the methods for constructing fault diagnosis models and using these models for prediction.

[0162] In this embodiment, QAR data and maintenance records of an airline's engines from January 2015 to July 2018 are used. The QAR data includes key parameters from four main systems: engine, flight control, hydraulic control, and environmental control, such as calculated airspeed (CAS), Mach number (M), altitude (ALT), fuel flow rate (FF), low-pressure turbine speed (N1), high-pressure turbine speed (N2), and exhaust gas temperature (EGT). The maintenance records include the times the engines were water-washed and the occurrence of various malfunctions. Water washing occurred a maximum of 54 times, spark plug replacements occurred 8 times, bird strike checks occurred 6 times, and fuel pump replacements occurred 3 times. Figure 13 A complete flowchart for constructing the fault diagnosis model of this invention is shown below. Figure 13 As shown, the specific steps are as follows:

[0163] Step 1: The preprocessing of QAR data mainly includes two parts: handling missing values ​​and extracting cruise points. The specific steps are as follows:

[0164] Step 1.1: Use the nearest neighbor imputation method to imput all missing values.

[0165] Step 1.2: Extract the cruise stability point.

[0166] Step 2: Optimize fault features by adding and filtering fault features to improve the accuracy of aero-engine fault diagnosis. Specific optimization steps are as follows:

[0167] Step 2.1: Since the current QAR data spans from January 2015 to July 2018 and does not include data from when the engine first entered service, this invention selects data from a period after water washing for baseline modeling.

[0168] Step 2.2: Baseline modeling is performed using a multiple linear method. The independent variables for the baseline are N1, TAT, M, and ALT, and the baseline includes EGT, N2, FF, T3, and T25.

[0169] Step 2.3: Use exponential feature augmentation, logarithmic feature augmentation, combined feature augmentation and difference feature augmentation to augment the dimensionality of five fault features: △FF, △N2, △EGT, △T3 and △T25. After augmentation, there are a total of 36 fault features.

[0170] Step 2.4: Use the nearest neighbor principal component analysis method to optimize features and select the optimal feature combination from 36 features.

[0171] Step 3: Develop an aero-engine fault diagnosis model based on refined KNN, based on feature optimization.

[0172] Step 4: Input the aircraft engine fault data to be diagnosed into the aircraft engine fault diagnosis model to predict the diagnosis results.

[0173] Step 4.1: Before conducting the diagnosis, it is necessary to perform sample equalization on the original aircraft engine fault data to be diagnosed.

[0174] Step 4.2: Perform data normalization processing on the balanced fault data.

[0175] Step 4.3: Based on sample equalization and data normalization, input the normalized fault features to be diagnosed into the aero-engine fault diagnosis model for diagnosis and prediction.

[0176] The following describes the aero-engine fault diagnosis model training device provided by the present invention. The aero-engine fault diagnosis model training device described below and the aero-engine fault diagnosis model training method described above can be referred to in correspondence. Figure 14 This is a schematic diagram of the structure of the aero-engine fault diagnosis model training device of the present invention, as shown below. Figure 14 As shown, the device includes:

[0177] The acquisition unit 1401 is used to acquire the deviation values ​​of the performance parameters of the aero-engine, including: exhaust temperature EGT, high-pressure turbine speed N2, and fuel flow rate FF;

[0178] The optimization unit 1402 is used to perform feature optimization on the deviation value of the performance parameter to obtain the optimized performance parameter deviation value;

[0179] The construction unit 1403 is used to construct an aero-engine fault diagnosis model based on the KNN algorithm, using the optimized performance parameter deviation value as input.

[0180] The present invention also provides an aircraft engine fault diagnosis device. Figure 15 This is a schematic diagram of the structure of an aircraft engine fault diagnosis device, such as... Figure 15 As shown, the device includes:

[0181] Acquisition unit 1501 is used to acquire the fault data of the aircraft engine to be diagnosed;

[0182] The prediction unit 1502 is used to input the aircraft engine fault data to be diagnosed into a pre-built aircraft engine fault diagnosis model for prediction and to obtain the diagnosis result.

[0183] The diagnostic results include: which type of fault the aircraft engine fault data to be diagnosed belongs to; and the aircraft engine fault diagnosis model is constructed using the method described in this invention.

[0184] Figure 16An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 16 As shown, the electronic device may include a processor 1610, a communications interface 1620, a memory 1630, and a communication bus 1640. The processor 1610, communications interface 1620, and memory 1630 communicate with each other via the communication bus 1640. The processor 1610 can call logical instructions from the memory 1630 to execute the aero-engine fault diagnosis model training method or the aero-engine fault diagnosis method.

[0185] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0186] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0187] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0188] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A training method for an aero-engine fault diagnosis model, characterized in that, include: Obtain the deviation values ​​of the performance parameters of the aero-engine, including: exhaust temperature EGT, high-pressure turbine speed N2, and fuel flow rate FF; The deviation values ​​of the performance parameters are subjected to feature optimization to obtain optimized performance parameter deviation values; Using the optimized performance parameter deviation values ​​as input, an aero-engine fault diagnosis model is constructed based on the KNN algorithm; The deviation values ​​for obtaining the performance parameters of the aero-engine include: obtaining the measured performance parameters of the aero-engine; performing baseline modeling based on a multivariate linear method to obtain the baseline performance parameters of the aero-engine, wherein the independent variables for the baseline modeling are low-pressure turbine speed N1, total atmospheric temperature TAT, Mach number M, and flight altitude ALT, and the established baseline includes EGT, N2, FF, high-pressure compressor outlet temperature T3, and low-pressure compressor outlet temperature T25; the baseline expression formula is as follows: in a ~ e These are the undetermined coefficients of the linear regression equation; Subtract the baseline performance parameter from the measured performance parameter to obtain the deviation value of the performance parameter; Before baseline modeling, the original performance parameters are preprocessed and used as the basis for baseline modeling. The pretreatment measures include: The nearest neighbor imputation method was used to fill in all missing values ​​in the original performance parameter data to obtain complete performance parameter data; Extract the performance parameter data of the aircraft cruise segment from the complete performance parameter data; Data from a period of time after water washing was selected from the performance parameter data of the aircraft cruise segment for baseline modeling; The step of performing feature optimization on the deviation values ​​of the performance parameters to obtain optimized performance parameter deviation values ​​includes: The deviation values ​​of the performance parameters are augmented using exponential feature augmentation, logarithmic feature augmentation, combined feature augmentation, and difference feature augmentation methods, respectively, to obtain the dimension-enhanced deviation values ​​of the performance parameters. The deviation values ​​of the performance parameters after dimensionality enhancement were screened using the nearest neighbor principal component analysis method to obtain the optimal deviation values ​​of the performance parameters. The optimal performance parameter deviation value is used as the basis for constructing the aero-engine fault diagnosis model; The method for constructing an aero-engine fault diagnosis model based on the KNN algorithm includes optimizing the number of neighboring points and the distance metric of the KNN algorithm, wherein the number of neighboring points is 1 and the distance metric is Euclidean distance.

2. A method for diagnosing aircraft engine faults, characterized in that, include: Acquire data on the aircraft engine faults to be diagnosed; The fault data of the aircraft engine to be diagnosed is input into a pre-built aircraft engine fault diagnosis model to obtain the diagnosis results; The diagnostic results include: what type of fault the aircraft engine fault data to be diagnosed belongs to; The aircraft engine fault diagnosis model is constructed using the method described in claim 1.

3. The aircraft engine fault diagnosis method according to claim 2, characterized in that, The acquisition of the aircraft engine fault data to be diagnosed includes: Acquire raw fault data of aircraft engines; The original fault data is augmented using a minority class oversampling method to obtain augmented fault data; The expanded fault data is normalized to obtain normalized fault data. The normalized fault data is used as the fault data of the aero-engine to be diagnosed.

4. A training device for an aircraft engine fault diagnosis model, characterized in that, To implement the method of claim 1, the method comprises: The acquisition unit is used to acquire the deviation values ​​of the performance parameters of the aero-engine, including: exhaust temperature EGT, high-pressure turbine speed N2, and fuel flow rate FF; An optimization unit is used to perform feature optimization on the deviation values ​​of the performance parameters to obtain optimized performance parameter deviation values. The construction unit is used to construct an aero-engine fault diagnosis model based on the KNN algorithm, using the optimized performance parameter deviation value as input.

5. An aircraft engine fault diagnosis device, characterized in that, To implement the method of claim 2, the method comprises: Acquisition unit, used to acquire fault data of the aircraft engine to be diagnosed; The prediction unit is used to input the aircraft engine fault data to be diagnosed into a pre-built aircraft engine fault diagnosis model for prediction and to obtain the diagnosis result. The diagnostic results include: what type of fault the aircraft engine fault data to be diagnosed belongs to.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-3.

Citation Information

Patent Citations

  • Performance offset-based aeroengine fault diagnosis method

    CN106503746A