CN112749764A - Aeroengine running state classification method based on QAR data - Google Patents

Abstract

The invention relates to the technical field of aero-engine operating state data analysis, in particular to an aero-engine operating state classification method based on QAR data, comprising: preprocessing the acquired QAR data; performing image processing on the preprocessed QAR data; The image-processed QAR data is put into a convolutional neural network for training to generate a classification model of the operating state of the aero-engine; the classification model is used to classify the fault state of the engine that has failed. The classification method provided by the present invention combines the image processing of the QAR data of the whole flight segment with the convolutional neural network to obtain a classification model, and then uses the classification model to classify the data for fault state. This classification method can achieve an accurate classification effect, which is helpful for the study of aero-engine status identification and fault diagnosis. Therefore, it can not only improve flight safety and reduce costs, but also is of great significance for aero-engine health management based on data-driven methods.

Description

A classification method of aero-engine operating state based on QAR data

technical field

The invention relates to the technical field of aero-engine operating state data analysis, in particular to a QAR data-based aero-engine operating state classification method.

Background technique

Aero-engine is the main power source during the flight of the aircraft, and its health and working state directly affect the safety of flight. The deterioration or failure of aero-engine performance will change its operating state, which in turn will lead to changes in various parameters of the engine during operation. The change of its operating parameters will be recorded in detail and truthfully by the aircraft marking configuration component QAR (Quick Access Recorder), which can continuously and completely reflect the actual state of the aircraft system in operation or the symptom signal of failure, and has easy access. , Simple maintenance, large data storage capacity, cheap on-board storage devices, and strong versatility. Therefore, QAR data is very important in monitoring the working state of aero-engines.

The existing methods of classifying fault states based on aero-engine simulation data can achieve high accuracy, but the simulation data may not reflect the distribution law of real operating data, and the engine data in real operating conditions contain a lot of noise, and different There is a strong nonlinear and coupling correlation between type parameters, and traditional data processing and classification methods are difficult to achieve good classification results. At the same time, most of the existing research on the flight data of aero-engines selects the data of the cruise stable point. However, the aircraft rarely has major faults during the cruise steady flight process, so it is impossible to comprehensively and accurately classify the fault state of the data. .

SUMMARY OF THE INVENTION

In order to solve the above-mentioned technical problem that the fault state or operating state of an aero-engine cannot be completely and accurately classified, the present invention provides a method for classifying the operating state of an aero-engine based on QAR data, which includes the following steps: S100. preprocessing the QAR data; S200, performing image processing on the preprocessed QAR data; S300, putting the image processing QAR data into a convolutional neural network for training to generate an aero-engine operating state classification model; S400, The classification model is used to classify the fault state of the failed engine.

On the basis of the above technical solution, further, in step S100, the preprocessing of the acquired QAR data includes outlier processing and missing value filling.

将原数据中的异常值或缺失值替换为X_m(n)。On the basis of the above technical solution, further, the processing method of the abnormal value processing and missing value filling is the mean value imputation method, that is, the nth number of the mth feature is set to be an abnormal value or a missing value, which is denoted as X _m (n), by formula

Replace outliers or missing values in the original data with X _m (n).

On the basis of the above technical solution, further, in step S200, the QAR data image processing process includes:

S210. Divide the original QAR data into different operating state categories according to the water washing records;

S220. Under the same operating state category, divide the data of each flight cycle into several segments according to the segment classification standard, including at least one of the take-off segment, the climb segment, the cruise segment, the descent segment and the landing segment;

S230, taking points at equal distances from each segment of data;

S240: Convert the data obtained by taking the points at equal distances into three-channel data.

On the basis of the above technical solution, further, in step S220, the division criteria of different flight segments are as follows:

Takeoff section: CAS>45&ALT<2000

Climb section: 2000<ALT<20000

Cruise segment: ALT＞20000&diff(ALT)<50

Landing stage: 2000<ALT<20000

Landing segment: CAS>45&ALT<2000

Among them, CAS is the calculated air speed, ALT is the flight altitude, and diff is the difference function.

On the basis of the above technical solution, further, the data of the take-off segment and the climb segment are in line with the requirements of the segment classification standard and are extracted from the QAR data of the first half of the entire flight cycle. The data of the segment meets the requirements of the flight segment classification standard and is extracted from the QAR data in the last half of the entire flight cycle.

On the basis of the above technical solution, further, in step S240, the data after the equidistant points are converted into three-channel data, including taking n equidistant data points from the take-off section and the climb section as the first channel. Data, take 2n equidistant data points from the cruise segment as the second channel data, and take n equidistant data points from the descent segment and the landing segment as the third channel data, so that the data scale after multi-channel data conversion is the same .

On the basis of the above technical solution, further, the three-channel data is set as the image data in which the number of row pixels is 2n, and the number of column pixels is the number of aero-engine features recorded in the original QAR data.

On the basis of the above technical solution, further, in step S300, the process of obtaining the classification model of the aero-engine operating state includes: standardizing the QAR data after image processing according to the following formula, so as to train through a convolutional neural network Standardized data to obtain aero-engine operating state classification model;

in,

On the basis of the above technical solution, further, the S300 further includes dividing the standardized data into a training set and a test set according to a certain proportion; the training set is used to train the standardized data through a convolutional neural network to obtain aero-engine operation. The state classification model, the test set is used to observe the model classification accuracy.

Compared with the prior art, the method for classifying the operating state of aero-engine based on QAR data provided by the present invention has the following advantages: image processing of the QAR data of the whole flight segment is combined with a convolutional neural network to obtain aviation The engine operating state classification model is used to classify the fault state of the QAR data. This classification method can achieve an accurate classification effect, which is helpful for the study of aero-engine status identification and fault diagnosis. Improve flight safety while reducing maintenance costs.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description These are some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

Fig. 1 is the step flow chart of the aero-engine operating state classification method based on QAR data provided by the present invention;

Figure 2 is a classification diagram of aero-engine state data;

Fig. 3 is the flow chart of aero-engine QAR data image processing;

Fig. 4 is a flowchart of image data input convolutional neural network training steps;

Fig. 5 is a schematic diagram of model classification accuracy change during training;

Figure 6 is a confusion matrix diagram for model classification.

Detailed ways

In order to make the purposes, 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 These are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

A method for classifying the operating state of an aero-engine based on QAR data provided by the present invention includes the following steps: S100, preprocessing the acquired QAR data; S200, performing image processing on the preprocessed QAR data; S300, converting the image The processed QAR data is put into a convolutional neural network for training to generate a classification model of the operating state of the aero-engine; S400, the classification model is used to classify the failure state of the engine that has failed.

In the specific implementation, the QAR data and maintenance records of an aero-engine are first introduced. Each QAR data file records the changes of more than 20 parameters in a complete flight cycle, including the flight altitude of the engine (Altitude, ALT), atmospheric Environmental parameters such as Total Air Temperature (TAT), performance parameters such as High Pressure Rotor Speed (N2), Exhaust Gas Temperature (EGT), and Variable Stator Vane, VSV) and other control parameters, while the maintenance record includes the time the engine was washed. When classifying QAR data, due to the difference in the frequency of data recorded by sensors in different parts of the aero-engine and the influence of the operating environment on the sensors, the QAR data files often contain various errors, and the existence of these errors will affect the effective information in the QAR data. extraction. Therefore, as shown in Figure 1, it is necessary to preprocess the QAR data first, and then use the preprocessed data for image processing to make it better suited to the convolutional neural network, and then input it into the convolutional neural network Training is used to generate an aero-engine operating state classification model, and the purpose of classifying the fault state of the faulty engine is achieved through the model.

The present invention provides a method for classifying the operating state of an aero-engine based on QAR data. The whole-segment QAR data is image-processed and combined with a convolutional neural network to obtain a classification model for the operating state of an aero-engine, and the classification model is used. Fault status classification for QAR data. This classification method can achieve an accurate classification effect, which is helpful for the study of aero-engine status identification and fault diagnosis. Improve flight safety while reducing maintenance costs.

Preferably, the preprocessing of the acquired QAR data in step S100 includes outlier processing and missing value filling.

During specific implementation, processing outliers and missing values in the QAR data can effectively reduce errors existing in data transmission, so as to avoid affecting subsequent monitoring and diagnosis.

将原数据中的异常值或缺失值替换为X_m(n)。Preferably, the processing method of the outlier processing and missing value filling is the mean value imputation method, that is, the nth number of the mth feature is set to be an outlier or missing value, denoted as X _m (n), through the formula

Replace outliers or missing values in the original data with X _m (n).

During the specific implementation, the outliers and missing values are processed by the mean value imputation method, and the outliers and missing values are replaced by the average value of the previous value and the latter value, so as to stabilize the data in a more effective range, to reduce errors.

Preferably, in step S200, the QAR data image processing process includes:

S210. Divide the original QAR data into different operating state categories according to the water washing records;

S220. Under the same operating state category, divide the data of each flight cycle into several segments according to the segment classification standard, including at least one of the take-off segment, the climb segment, the cruise segment, the descent segment and the landing segment;

S230, taking points at equal distances from each segment of data;

S240: Convert the data obtained by taking the points at equal distances into three-channel data.

In specific implementation, due to the fact that there are few aero-engine failure samples in the real flight process, but the service time is prolonged, the performance degradation of the aero-engine will also lead to changes in the data distribution, and washing the engine can restore the performance degradation of the engine due to fouling. However, the performance of the engine after water washing will still decline to a certain extent compared with the performance of the engine after the previous water washing. Therefore, the different operating states of the aero-engine can be distinguished by the water washing time. As shown in Figure 2, the original QAR data is divided into four different operating states according to the five times of water washing maintenance records.

Therefore, when image processing of QAR data, it is necessary to divide the original QAR data into different operating state categories according to the washing records, and under the same operating state category, divide the data of each flight cycle into several segments according to the division criteria of the flight segment. Section, as a preferred solution, is divided into five basic stages of each normal flight of the aircraft, namely take-off section, climb section, cruise section, descent section and landing section. Secondly, the points of each segment of data are equidistant, the purpose is to reduce the amount of data on the premise of ensuring that there is not much information loss, and at the same time to ensure that the data size of the three channels is the same. Finally, the data after equidistant points are converted into three-channel data, that is, the data is imaged as a sample of the neural network.

Preferably, in step S220, the division criteria for different flight segments are as follows:

Takeoff section: CAS>45&ALT<2000

Climb section: 2000<ALT<20000

Cruise segment: ALT＞20000&diff(ALT)<50

Landing stage: 2000<ALT<20000

Landing segment: CAS>45&ALT<2000

Among them, CAS is the calculated air speed, ALT is the flight altitude, and diff is the difference function.

In specific implementation, since the flight stage of the aero-engine in each flight cycle is different and the engine operating state is different, the data of each flight cycle is divided into five flight segments according to the air speed and flight height. Using this classification standard is simple and easy to implement in the operation process, and does not require too many parameter interventions.

Preferably, the data of the take-off segment and the climbing segment meet the requirements of the flight segment classification standard and are extracted from the QAR data of the first half of the entire flight cycle, and the data of the landing segment and the landing segment meet the flight segment classification requirements. Standard requirements and extracted in the last half of the QAR data of the entire flight cycle.

Preferably, in step S240, converting the data after equidistant points into three-channel data, including taking n equidistant data points from the takeoff section and the climb section as the first channel data, and taking 2n data points from the cruise section The equidistant data points are used as the second channel data, and n equidistant data points are taken from each of the descending segment and the landing segment as the third channel data, so that the data scales of the multi-channel data after conversion are the same.

During the specific implementation, based on the data obtained after taking the points at equal distances, on the one hand, because the working states of the take-off segment and the climbing segment are similar, that is, the aero-engine generates a large thrust, the engine in the cruise segment is in a stable working state, and the working states of the descending segment and the landing segment are in a stable working state. Similarity means that the thrust is small, so the flight stages with similar working states can be regarded as the same channel. On the other hand, in order to make the data scale of the multi-channel data conversion the same, n equidistant data points are taken from each of the take-off segment and the climb segment. As the first channel data; take 2n equidistant data points from the cruise segment as the second channel data; take n equidistant data points from the descent segment and the landing segment as the third channel data. In this way, the original QAR data is transformed into three-channel data as a neural network sample, and the flow chart of the transformation is shown in Figure 3.

Preferably, the three-channel data is set as the image data in which the number of row pixels is 2n, and the number of column pixels is the number of aero-engine features recorded in the original QAR data.

In specific implementation, the three-channel data is set as the number of row pixels is 2n, and the number of column pixels is the image data of the number of aero-engine features recorded in the original QAR data, which is convenient for subsequent input into the convolutional neural network.

Preferably, in step S300, the process of obtaining the aero-engine operating state classification model includes: standardizing the QAR data after image processing according to the following formula, so as to obtain the aero-engine operating state by training the standardized data through a convolutional neural network classification model;

in,

In the specific implementation, as shown in Figure 4, the training process of the convolutional neural network is shown in Figure 4. The QAR data after image processing is used as the sample of the neural network for data standardization, and its function is to eliminate the influence of different parameter dimensions in the machine learning process. Then put it into the convolutional neural network, and after the training of the neural network, the above-mentioned high-accuracy classification model of the aero-engine operating state can be obtained.

Preferably, the S300 further includes dividing the standardized data into a training set and a test set according to a certain proportion; the training set is used to obtain an aero-engine operating state classification model by training the standardized data through a convolutional neural network, and the test set Used to observe the model classification accuracy.

During specific implementation, the standardized data may be divided into a training set and a test set according to a certain proportion, wherein the data of the training set is more than the data of the test set. Then put the data of the training set into the convolutional neural network to train the aero-engine operating state classification model, and finally use the test set to test the accuracy of the model classification. Through this method step, the model can be well tested. As an embodiment, the data of the training set and the test set are divided according to the ratio of 3:1, and the test set is input into the model, as shown in Figure 5, which shows that in the training process of the neural network, the classification The accuracy of the model is increasing, and the final classification accuracy can reach more than 98%; as shown in Figure 6, it shows the confusion of model classification between different operating states, and it can be found that the model has a good classification accuracy for different operating states. Therefore, a good aero-engine operating state classification model can be obtained by the method provided by the present invention, which can better classify the fault state or operating state of the aero-engine, and the high classification accuracy means that the fault can be accurately located, and the Effectively reduce the sudden failure of the engine.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, but not 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: The technical solutions described in the foregoing embodiments can still be modified, or some or all of the technical features thereof can be equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present invention. scope.

Claims (10)

1. An aeroengine operating state classification method based on QAR data is characterized by comprising the following steps:

s100, preprocessing the acquired QAR data;

s200, carrying out imaging processing on the preprocessed QAR data;

s300, putting the QAR data subjected to imaging processing into a convolutional neural network for training to generate an aeroengine operation state classification model;

and S400, adopting the classification model to classify the fault state of the engine with the fault.

2. The QAR data-based aircraft engine operating condition classification method according to claim 1, wherein: the preprocessing of the QAR data acquired in step S100 includes outlier processing and missing value padding.

3. The QAR data-based aircraft engine operating condition classification method according to claim 2, wherein the QAR data is based on a relationship between the operating conditions of the aircraft engine and the operating conditions of the aircraft engineCharacterized in that: the processing method of abnormal value processing and missing value filling is an average value filling method, namely the nth number of the m-th characteristic is an abnormal value or a missing value and is marked as X_m(n) by the formula

Replacing abnormal values or missing values in the original data with X_m(n)。

4. The QAR data-based aircraft engine operating condition classification method according to claim 1, wherein: in step S200, the QAR data imaging process includes:

s210, dividing original QAR data into different operation state categories according to washing records;

s220, under the same operation state type, dividing the flight cycle data of each time into a plurality of sections according to a flight section division standard, wherein each section at least comprises one of a takeoff section, a climbing section, a cruise section, a descent section and a landing section;

s230, equally dividing the data of each section into points;

and S240, converting the data with the equidistant points into three-channel data.

5. The QAR data-based aircraft engine operating condition classification method according to claim 4, wherein: in step S220, the division criteria for different legs are as follows:

a takeoff section: CAS > 45& ALT <2000

A climbing section: 2000< ALT <20000

And (3) a cruise section: ALT > 20000& diff (ALT) <50

A landing section: 2000< ALT <20000

Landing stage: CAS > 45& ALT <2000

Where CAS is the calculated air velocity, ALT is the fly height, and diff is the difference function.

6. The QAR data-based aircraft engine operating condition classification method according to claim 5, wherein: the data of the takeoff segment and the climb segment are extracted from QAR data which meet the requirements of the flight segment division standard and are one half of the data in front of the whole flight cycle, and the data of the landing segment are extracted from QAR data which meet the requirements of the flight segment division standard and are one half of the data in back of the whole flight cycle.

7. The QAR data-based aircraft engine operating condition classification method according to claim 4, wherein: in step S240, the data after equidistant point extraction is converted into three-channel data, including taking n equidistant data points from the takeoff segment and the climb segment as first channel data, taking 2n equidistant data points from the cruise segment as second channel data, and taking n equidistant data points from the descent segment and the landing segment as third channel data, so that the data scale after multi-channel data conversion is the same.

8. The QAR data-based aircraft engine operating condition classification method according to claim 7, wherein: and setting the three-channel data as imaging data with the row pixel number of 2n and the column pixel number of the aeroengine characteristic number recorded by the original QAR data.

9. The QAR data-based aircraft engine operating condition classification method according to any one of claims 1 to 8, wherein: in step S300, the process of obtaining the classification model of the operating state of the aircraft engine includes: carrying out data standardization on the QAR data subjected to imaging processing according to the following formula so as to obtain an aircraft engine operation state classification model by training standardized data through a convolutional neural network;

wherein,

10. the QAR data-based aircraft engine operating condition classification method according to claim 9, wherein: the S300 further comprises dividing the standardized data into a training set and a testing set according to a certain proportion; the training set is used for training standardized data through a convolutional neural network to obtain an aero-engine running state classification model, and the testing set is used for observing the classification accuracy of the model.

Images