Automatic evaluation method for bird strike damage of aircraft engine

Abstract

The invention relates to the technical field of aircraft engine health monitoring and fault diagnosis, in particular to a bird strike damage automatic evaluation method for a CFM56-7B aircraft engine. According to the method, key operation parameters are screened through the working principle of an engine, and a double-layer cross validation architecture of a double-engine difference feature layer and a single-engine deviation feature layer is constructed; instantaneous disturbance is captured by calculating the difference value change peak value of the same parameters of the two engines, and single-engine internal parameter logic departure and serious departure events are identified by utilizing high-correlation parameters of the engine to correlation coefficients; and training an intelligent classification model based on the features, and finally outputting a three-level linkage decision-making system of a parameter abnormity level, a damage risk level and a maintenance suggestion. According to the method, tiny and transient parameter anomalies caused by bird strike can be accurately recognized, missing detection and false alarm are remarkably reduced, the damage degree is rapidly and automatically evaluated, reliable support is provided for bird strike engine decision making, and flight safety is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of aircraft engine health monitoring and fault diagnosis, and particularly relates to an automatic evaluation method for bird strike damage of a CFM56-7B engine. BACKGROUND

[0002] Bird strike of an aero-engine is a common external damage event in flight operation, which often leads to in-flight shutdown, interrupted take-off and other serious consequences, and directly threatens flight safety. Most of such events occur during the take-off and approach phases, when the crew has heavy operating load and mainly concentrates on aircraft control and communication, and it is difficult to effectively detect and evaluate the subtle abnormalities and internal damage of engine parameters. In addition, some bird strike traces may become less obvious due to rain washing or continuous flight, further increasing the risk of missed detection in ground inspection.

[0003] Currently, the aircraft maintenance manual (AMM) has a fuzzy zone for the definition of engine parameters after bird strike. For example, the manual states that short-term fluctuations in engine parameters caused by bird strike can be considered as normal phenomena, and in the case of normal parameters and no visual material loss of fan blades, the internal channel hole probe inspection is allowed to be delayed for a certain period. This provision brings significant decision-making difficulties in actual operation: on the one hand, the professional ability and experience of the on-duty engineer vary, making it difficult to accurately determine whether the "short-term fluctuations" hide internal damage; on the other hand, to avoid safety risks, the "one-size-fits-all" immediate stop for hole probe inspection will bring great pressure and unnecessary delay loss to flight operation. Therefore, in view of the above deficiencies, an automatic evaluation method for bird strike damage of an aircraft engine is proposed. SUMMARY

[0004] (I) Technical problems to be solved In view of the deficiencies in the prior art, the present application provides an automatic evaluation method for bird strike damage of an aircraft engine, which solves the problems of unclear definition of engine parameters after bird strike, which makes it difficult to accurately determine whether the parameter fluctuations correspond to bird strike events, and the problem of non-uniform evaluation standards for bird strike events.

[0005] (II) Technical solutions To solve the above problems, the present application provides an automatic evaluation method for bird strike damage of an aircraft engine, comprising: Step S1: collecting historical operation data of the target aircraft engine, selecting operation parameters related to bird strike damage and pre-processing the data; Step S2: calculating the difference between the same parameters of the left and right engines to obtain a dual-engine difference benchmark, and calculating the dual-engine difference deviation for the operation parameters selected in step S1; Step S3: Define single-engine deviation parameter pairs in the screened operating parameters, and calculate the correlation coefficient of the single-engine high correlation parameter pairs; Step S4: For the screened parameters, calculate the mutual correlation of the parameter pairs according to the left and right engine same parameter double-engine difference deviation calculated in step S2 and the high correlation parameter pairs in step S3, identify the historical events of double-engine parameter deviation and single-engine parameter deviation; Step S5: Determine the severity of the double-engine parameter difference feature, the single-engine parameter deviation feature and the historical bird strike event, and label the label according to the severity; Step S6: According to the labeling result in step S5, combine the historical engine bird strike damage situation to re-label to form a data set; Step S7: Use the data set obtained in step S6 to intelligently train the model, and verify the model training effect; Step S8: Input the measured data of the target aircraft engine into the trained model to evaluate the bird strike damage.

[0006] Further, the step S1 comprises: S101: Eliminate outliers in the collected data; S102: Perform double-engine noise reduction processing on each operating parameter.

[0007] Further, the operating parameters include engine low-pressure rotor speed (N1), engine high-pressure rotor speed (N2), engine exhaust gas temperature (EGT), engine fuel flow (FF), engine high-pressure compressor outlet pressure (PS3), engine vibration value (VIB), engine throttle lever angle (TRA) and the like.

[0008] Further, the step S3 defines the deviation parameter pairs according to the inherent logical relationship between the related operating parameters in the process of aircraft engine operation.

[0009] Further, in the step S4, in the historical information of the related operating parameters, the time point where the parameter change in the deviation parameter pair does not conform to the inherent logic is determined as a deviation event; the time period in which two deviation events occur within a specified time is defined as a serious deviation event.

[0010] Further, in the step S4, the deviation event is determined according to the correlation degree of different time periods, and the correlation degree calculation formula is: ; Wherein, : The correlation coefficient of the ith comparison sequence and the reference sequence at point k; : The value of the reference sequence at point k; : the value of the i-th comparison sequence at the k-th point. The comparison sequence represents the sequence to be evaluated, representing the parameters in the bird strike event; : the absolute difference of the reference sequence and the comparison sequence at the k-th point; : the minimum absolute difference of all points of all comparison sequences; : the maximum absolute difference of all points of all comparison sequences.

[0011] Further, the engine bird strike damage condition label in the step S6 includes a severe bird strike, a bird strike, and a non-bird strike.

[0012] Further, the output result of the intelligent learning model in the steps S6-S8 includes a parameter abnormality level, a damage risk level, and a maintenance suggestion.

[0013] (Three) beneficial effects The aircraft engine bird strike damage automatic evaluation method provided by the present application screens the operating parameters related to bird strike damage, calculates the difference characteristics of the same parameters of the two engines and the deviation characteristics of the high correlation parameters of the single engine in the running process, establishes the correlation between the difference characteristics of the two engines and the deviation parameters of the single engine and the bird strike damage of the engine, and inputs the established correlation into the model for learning and training. It can accurately identify the parameter change characteristics caused by bird strikes, determine the bird strike risk in different operating states, and reduce the missed detection rate; it can accurately and quickly judge the degree of bird strike damage and output the judgment result in time, so as to realize scientific and quantitative evaluation of the damage degree of the engine, provide reliable and consistent technical basis for maintenance release decision, and find the best balance point between ensuring flight safety and improving operating efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 The flowchart of the aircraft engine bird strike damage automatic evaluation method of the present application. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0016] In the description of this invention, it is necessary to understand that the orientation or positional relationship indicated by terms such as "upper", "lower", "inner", "outer", "top", and "bottom" are based on the orientation or positional relationship shown in the accompanying drawings. The purpose is only to facilitate the description of this invention and to simplify the description. It is not intended to indicate or imply that the component referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0017] like Figure 1 As shown, this invention provides an automatic evaluation method for bird strike damage to aircraft engines, specifically including: Step S1: Collect historical operating data of the target aircraft engine and select operating parameters related to bird strike damage based on the engine's operating principle. To effectively identify and judge the changes in aircraft operating parameters caused by bird strike damage, it is necessary to use the historical operating data of the target aircraft engine as a basis to select operating parameters with a high correlation to bird strike damage as relevant operating parameters. These relevant operating parameters will be used as the benchmark for judging the bird strike damage status in subsequent assessments. When the aircraft engines are put into service, they are arranged in pairs, with each pair mounted on one of the wings. When collecting historical operating data, data from both the left and right engines must be collected simultaneously to ensure data accuracy.

[0018] Step S1 includes: Outliers are removed from the collected data. The historical operating data of the aircraft engine is usually obtained from the Quick Access Recorder (QAR) of the aircraft where the target engine is located. It contains all historical data of the operating parameters of the left and right engines. At this time, there are some error data due to measurement errors and other reasons. In order to ensure the accuracy of subsequent calculation and judgment results, the data needs to be preprocessed to remove errors and outliers.

[0019] During the preprocessing process, a separate parameter column x is created for each running parameter. j Then calculate the mean μ of each parameter column. x and standard deviation σ x If a certain parameter x in the sequence j The difference between the value and the mean is greater than three times the standard value, i.e., |x j −μ x If |>3σx, then xj is considered an outlier and is removed.

[0020] The process of removing outliers can be represented in code as follows: """ self.sigma_threshold = sigma_threshold self.scaler = StandardScaler() def remove_outliers_3sigma(self, df, columns): """ Where df: DataFrame, original data; columns: list of column names that need to be processed.

[0021] For a certain time point t, if the value xt at t is missing, the value is completed by interpolation method for missing, set the time point t before the time t0, after the time t1, then the interpolation method completion formula is: .

[0022] After preprocessing, the effective historical data set of each operating parameter is obtained.

[0023] After processing, the staff selects parameters associated with the bird strike height according to the working principle of the engine as the analysis parameters of the model of the application. Usually, these parameters include engine low-pressure rotor speed (N1), engine high-pressure rotor speed (N2), engine exhaust gas temperature (EGT), engine fuel flow (FF), engine high-pressure compressor outlet pressure (PS3), engine vibration value (VIB), engine throttle lever angle (TRA), etc.

[0024] Further, by comparing the deviation of the sequences of the above parameters at the same time point in the bird strike event, the closer the deviation form is, the higher the correlation between the two parameters is.

[0025] The two operating parameter sequences to be compared are defined as reference sequence x0 and comparison sequence x i In order to eliminate the dimension effect, the data in x0 and x i can be normalized, and the normalization formula is: Where μ x is the mean of the sequence, and σ x is the standard deviation of the sequence. Usually, the reference sequence x0 is the bird strike damage, and the comparison sequence x i is the operating parameter sequence to be tested.

[0026] Substitute the two sequences x0 and x i into the correlation calculation formula, and the correlation calculation formula is:

[0027] Where, At point k, the correlation coefficient between the i-th comparison sequence and the reference sequence; The value of the reference sequence at point k; The value of the i-th comparison sequence at point k. The comparison sequence represents the sequence to be evaluated, characterizing the parameters in the bird strike event; The absolute difference between the reference sequence and the comparison sequence at point k; The minimum absolute difference across all points in all comparison sequences; : The maximum absolute difference across all points in all comparison sequences.

[0028] S103: Select the running parameters with a correlation degree greater than 0.95 as the relevant running parameters.

[0029] Step S2: For the operating parameters selected in Step S1, calculate the difference between the same parameters of the left and right engines to obtain the dual-engine difference benchmark, and calculate the dual-engine difference deviation. For the preprocessed data, calculate the normal benchmark ΔP of the difference between the same operating parameters of the left and right engines. base And the real-time difference ΔP, calculate the real-time difference ΔP and the benchmark difference ΔP. base The difference between them yields the parameter deviation ΔP. dev Parameter deviation ΔP dev This represents the difference between the current left and right engine parameter differences and the baseline difference. The process of obtaining the engine difference characteristics is represented by the following code: # Calculate the deviation increment delta_p_dev = self.calculate_parameter_deviation(df, param) # Extracting statistical features features[f'delta_dev_{param}_max'] = np.max(delta_p_dev) features[f'delta_dev_{param}_mean'] = np.mean(delta_p_dev) features[f'delta_dev_{param}_std'] = np.std(delta_p_dev) features[f'delta_dev_{param}_peak'] = np.max(np.abs(delta_p_dev)) # Extracting temporal features: the number of times the threshold is exceeded if param == 'N1': threshold = 2.0 elif param == 'PS3': threshold = 20 elif param == 'EGT': threshold = 18 else: threshold = np.mean(delta_p_dev) + 2 * np.std(delta_p_dev) exceed_count = np.sum(delta_p_dev > threshold) features[f'delta_dev_{param}_exceed_count'] = exceed_count Step S3: Define single-engine deviation parameter pairs in the screened operating parameters, and calculate the correlation coefficient of the single-engine deviation parameter pairs. According to the working principle of the engine, there is a strong positive correlation between some operating parameters during the flight of the aircraft, so the change trend of the two parameters is highly similar. When the engine is hit by a bird, it may cause deformation of the compressor blades, flow field turbulence, and normal aerodynamic logic is destroyed, thus destroying the consistency of the change trend between two parameters with strong correlation, that is, indicating that a single-engine parameter deviation has occurred at this time. Under normal circumstances, an increase in fuel flow (FF) should lead to an increase in high-pressure compressor outlet pressure (PS3) and exhaust gas temperature (EGT); for example, an increase in throttle lever angle (TRA) will cause the engine speed (N1 / N2) to rise. Therefore, fuel flow can form a deviation parameter pair with high-pressure compressor outlet pressure (PS3) and exhaust gas temperature (EGT), respectively, and the throttle lever angle forms a deviation parameter pair with the engine speed. Common deviation pairs include (FF, PS3), (FF, EGT), (TRA, N1 / N2), (EGT, N1 / N2), etc.

[0030] In the actual training process, assuming that the parameter pair is (x, y), where x and y are two parameters in a set of deviation parameter pairs, the Pearson correlation coefficient of x and y in window W is calculated, and the length of window W is generally 5 seconds (also 1 second), which can be adjusted according to the needs of the work. The formula for calculating the Pearson correlation coefficient is: , where r xy (W) is the Pearson correlation coefficient of parameters x and y in window W; xi and y i are the i-th point parameter values of x and y within the window, respectively; are the mean of x and y within the window, respectively; W is the window size. The calculated Pearson correlation coefficient describes the correlation between two parameters.

[0031] The code for calculating the Pearson correlation coefficient of the sliding window is as follows: pearson_corr = self.calculate_sliding_correlation(series1, series2) features[f'{param1}_{param2}_pearson_min'] = np.min(pearson_corr) features[f'{param1}_{param2}_pearson_mean'] = np.mean(pearson_corr) features[f'{param1}_{param2}_pearson_std'] = np.std(pearson_corr).

[0032] Step S4: For the filtered parameters, calculate the cross-correlation of the parameter pair according to the left and right engine same parameter difference deviation calculated in step S2 and the high correlation parameter pair in step S3, and identify the historical events of double-engine parameter deviation and single-engine parameter deviation. Generally, parameter pairs with a calculation result greater than 0.98 are considered to be in normal operation state. According to the corresponding relationship between different parameter pairs and different working requirements, a preset threshold θ is set for the Pearson correlation coefficient of different parameter pairs. When the calculated result is less than the threshold θ, it is judged that a parameter deviation occurs. Severe bird strike events during flight will cause the operating parameters of the engine to have a secondary parameter fluctuation deviation feature. Therefore, the time period in which two deviation events occur within a specified time is defined as a severe deviation event. Generally, if two deviation events occur within 300s, it is judged that a severe parameter deviation event occurs within this time period, which should be defined as a severe damage, and immediate on-site inspection is recommended. According to the working environment and working requirements of different parameters during the flight of the aircraft, the normal operation range of each parameter is set. When the parameter deviation exceeds the normal operation range, it is judged that the parameter difference between the two engines is too large, and there is a bird strike risk.

[0033] The process of detecting deviation events and severe deviation time is represented by the following code: Detecting deviation events """ if param_pair == ('TRA', 'N1'): threshold = 0.95 # High-risk threshold elif param_pair == ('EGT', 'PS3'): threshold = 0.93 # High-risk threshold else: threshold = 0.90 # Default threshold The test showed a significant deviation (two deviations within a short period of time). severe_divergence = 0 for i in range(len(pearson_corr) - 1): if pearson_corr[i] < threshold and pearson_corr[i+1] < threshold: severe_divergence += 1 features[f'{param1}_{param2}_severe_divergence']=severe_divergence Step S5: Determine the characteristics of dual-parameter differences, single-parameter deviations, and the severity of single-parameter deviations, and label them according to the severity. Based on the identification results in steps S2 and S4, and combined with the specific circumstances of dual-parameter differences and historical deviation events, determine the parameter anomaly level. Typically, parameter anomaly levels include low, medium, and high anomaly levels. Staff use these three levels as labels and annotate the data at each time point according to the judgment results. Generally, the more frequent and severe the dual-parameter difference exceedances and single-parameter deviations occur within the same time point, the higher the corresponding anomaly level. The frequency and severity thresholds of dual-parameter difference exceedances and single-parameter deviation events corresponding to each anomaly level can be adjusted according to the actual model results.

[0034] Step S6: Based on the annotation results in Step S5, and combined with historical engine bird strike damage data, the dataset is re-annotated to form a dataset. After annotating the abnormality levels of the parameters, staff query historical maintenance records to determine the actual damage situation. Based on the damage situation, it is divided into three categories from low to high: severe bird strike, bird strike, and no bird strike. The corresponding labels for these three damage situations are high risk, medium risk, and low risk, respectively. After determining the damage situation at each time point, staff re-annotate the data in the dataset with the labels corresponding to the damage situation.

[0035] Step S7: Use the data set obtained in step S6 to intelligently train the model and verify the model training effect. Generally, the model used is preferably XGBoost or 1D-CNN, and in order to improve the training effect, the input feature dimension in the data set is usually not less than 8 dimensions, including engine low-pressure rotor speed N1, engine high-pressure rotor speed N2, engine exhaust gas temperature EGT, engine fuel flow FF, high-pressure compressor outlet pressure PS3, throttle lever angle TRA, fuel-air ratio FUELAIR RATIO, vibration value VIB, and stall index STALL.

[0036] The expression of the training data is: , Wherein, xi represents the i-th training sample (corresponding to the time point i or the data point i), which is an N-dimensional feature vector; Each element in the vector represents the j-th feature value of the i-th sample. For example, N1, N2, EGT, FF, PS3, TRA, FUEL AIR RATIO, PT2, VIB, and STALL in the QAR data; T represents the total number of training samples.

[0037] The mapping relationship of the model is: Wherein, θ is the parameter of the model (such as neural network weight); xi is the input sample (i.e. the feature vector in the above training data); Rstrike is the output of the model, representing the quantized damage risk level; f θ is a function approximator that learns a nonlinear mapping from input features to risk levels; In the training process, 80% of the data in the data set is used as the training set of the model to input the model for training; the remaining 20% of the data is used as the test question to test the learning achievement of the trained model. In actual work, the data proportion of the training set and the test set can be adjusted as needed.

[0038] Step S8: Input the to-be-measured data of the target aircraft engine into the trained model to evaluate the bird strike damage. After the model is trained, the corresponding parameters of the daily flight data are input into the trained model, the model performs bird strike identification and damage evaluation on the daily flight data, and outputs the results.

[0039] The output result of the intelligent learning model in S6-S8 includes a parameter anomaly level, a damage risk level, and a maintenance suggestion. The maintenance suggestion corresponding to the combination of the data anomaly level and the risk level is as follows: High data anomaly level → high risk level → suggest immediate hole-probe inspection; Medium anomaly level → medium risk level → suggest normal hole-probe release according to AMM parameters; Low anomaly level → low risk level → continue monitoring, no special treatment is needed.

[0040] Finally, it should be noted that the above description is only a preferred embodiment of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent replacements to some technical features, as long as they are within the spirit and principles of the present application. Any modification, equivalent replacement, improvement, etc. made within the scope of the present application should be included in the protection scope of the present application.

Claims

1. An automatic evaluation method for bird strike damage to aircraft engines, characterized in that, include: Step S1: Collect historical operating data of the target aircraft engine, select operating parameters related to bird strike damage, and preprocess the data; Step S2: For the operating parameters obtained in step S1, calculate the difference between the same parameters of the left and right engines to obtain the dual-engine difference benchmark, and calculate the dual-engine difference deviation. Step S3: Define single-shot divergence parameter pairs in the filtered operating parameters, and calculate the correlation coefficient of single-shot divergence parameter pairs; Step S4: For the filtered parameters, calculate the cross-correlation of the parameter pair based on the difference deviation between the left and right engines with the same parameters calculated in Step S2 and the highly correlated parameter pair in Step S3, and identify historical events of deviation between the parameters of the two engines and deviation of the parameters of a single engine. Step S5: Determine the severity of dual-shot parameter difference characteristics, single-shot parameter deviation characteristics, and single-shot parameter deviation historical events, and label them according to the severity. Step S6: Based on the annotation results in Step S5, and combined with historical engine bird strike damage data, annotate again to form a dataset; Step S7: Use the dataset obtained in step S6 to perform intelligent training on the model and verify the model training effect; Step S8: Input the target aircraft engine's measurement data into the trained model to evaluate the bird strike damage.

2. The automatic evaluation method for bird strike damage to aircraft engines according to claim 1, characterized in that, Step S1 includes: S101: Remove outliers from the collected data; S102: Perform dual-transmission noise reduction processing on various operating parameters.

3. The automatic evaluation method for bird strike damage to aircraft engines according to claim 1 or 2, characterized in that, The operating parameters include engine low-pressure rotor speed (N1), engine high-pressure rotor speed (N2), engine exhaust temperature (EGT), engine fuel flow rate (FF), engine high-pressure compressor outlet pressure (PS3), engine vibration value (VIB), and engine throttle lever angle (TRA).

4. The automatic evaluation method for bird strike damage to aircraft engines according to claim 3, characterized in that, In step S1, relevant operating parameters are selected based on the correlation between each parameter and the bird strike event. The correlation calculation formula is as follows: ; in, At point k, the correlation coefficient between the i-th comparison sequence and the reference sequence; The value of the reference sequence at point k; The value of the i-th comparison sequence at point k. The comparison sequence represents the sequence to be evaluated, characterizing the parameters in the bird strike event; The absolute difference between the reference sequence and the comparison sequence at point k; The minimum absolute difference across all points in all comparison sequences; : The maximum absolute difference across all points in all comparison sequences.

5. The automatic evaluation method for bird strike damage to aircraft engines according to claim 1, characterized in that, In step S3, divergent parameter pairs are defined based on the inherent logical relationship between relevant operating parameters during the operation of the aircraft engine.

6. The automatic evaluation method for bird strike damage to aircraft engines according to claim 5, characterized in that, In step S4, in the historical information of relevant operating parameters, the time point in which the parameter changes within the deviation parameter pair do not conform to the inherent logic is judged as a deviation event; the time period in which two deviation events occur within a specified time is defined as a serious deviation event.

7. The automatic evaluation method for bird strike damage to aircraft engines according to claim 1, characterized in that, In step S6, the engine bird strike damage status label includes severe bird strike, bird strike, and non-bird strike.

8. The automatic evaluation method for bird strike damage to aircraft engines according to claim 1, characterized in that, The output results of the intelligent learning model in S6-S8 include parameter anomaly level, damage risk level, and maintenance recommendations.

Images