A method for early warning of aircraft failures

Abstract

The application discloses an aircraft fault early warning method, and relates to the technical field of fault early warning.The method comprises the following steps: performing multi-level health index division on original flight data, performing data fusion and collection from bottom to top, and performing health state evaluation in a hierarchical manner to obtain a multi-level characteristic parameter spectrum and an overall system health state evaluation result; performing early warning grade determination on the system health state of the aircraft; acquiring information according to a structure manual, a wiring manual and a troubleshooting manual to which the aircraft belongs; constructing a basic graph structure of a fault transmission logic; correlating and fusing a fault propagation path according to acquired onboard multi-source information, an initial fault transmission characteristic graph and hierarchical fault early warning information; performing node mapping and diagnosis constraint condition construction on an obtained system failure state directed graph; performing priority sorting and propagation path tracing on fault causes; and quickly locating a cross-system coupling fault source.The application can improve the aircraft fault early warning efficiency.

Description

Technical Field

[0001] This invention relates to the field of video interactive control technology, and in particular to an aircraft fault early warning method. Background Technology

[0002] In the field of civil aircraft operation and support, early warning of aircraft malfunctions and rapid troubleshooting of flight routes are key means to ensure flight safety, improve aircraft utilization, shorten airport turnaround time, and reduce maintenance costs.

[0003] Existing aircraft early warning and troubleshooting technologies still have significant shortcomings: most only monitor a single system individually, ignoring the interrelationships between parameters and the cascading transmission of faults between systems, making it difficult to accurately locate complex coupled faults; conventional intelligent algorithms are mostly black-box operations, only outputting abnormal results without explaining the cause and propagation path of the fault, making it impossible for maintenance personnel to trace the reasoning process, thus limiting their practical troubleshooting reference value; traditional methods struggle to automatically correct judgment criteria in scenarios with few fault samples, easily leading to false alarms and missed alarms; moreover, existing early warning, mapping, diagnosis, and location functions are independent of each other, failing to form a complete closed loop from data processing, health assessment, and abnormal early warning to fault location and route troubleshooting, resulting in excessive reliance on maintenance personnel experience, long troubleshooting times, and extended aircraft turnaround times, all of which require further improvement. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, the present invention provides an aircraft fault early warning method.

[0005] This invention provides an aircraft malfunction early warning method, the method comprising:

[0006] Step S1: Obtain the aircraft's raw flight data. Based on the raw flight data, perform multi-level health indicator classification, bottom-up data fusion and summarization, and hierarchical health status assessment to comprehensively obtain the multi-level characteristic parameter spectrum and the overall system health status assessment results.

[0007] Step S2: Based on the multi-level feature parameter spectrum and the whole system health status assessment results, determine the warning level of the aircraft's system health status, obtain graded fault warning information, acquire and construct the basic graph structure of the fault transmission logic according to the structural manual, wiring manual and troubleshooting manual of the aircraft, obtain the initial fault transmission feature graph, acquire airborne multi-source information, and perform correlation and fusion of the fault propagation path according to the airborne multi-source information, the initial fault transmission feature graph and the graded fault warning information to obtain the directed graph of system failure status;

[0008] Step S3: Node mapping and diagnostic constraint construction are performed on the directed graph of the system failure state to obtain fault diagnosis constraints. Based on the fault diagnosis constraints and the directed graph of the system failure state, the causes of the fault are prioritized and the propagation path is traced to obtain a list of root causes and a fault propagation causal chain. Based on the list of root causes and the fault propagation causal chain, the source of the cross-system coupled fault is quickly located to obtain the fault location result.

[0009] Preferably, the original flight data is divided into four flight phases: takeoff and climb, cruise and steady-state, descent and approach, and landing and taxiing, to obtain takeoff phase data, cruise phase data, descent phase data, and landing phase data.

[0010] The takeoff, cruise, descent, and landing data are classified and statistically analyzed to determine the correlation between abnormal fluctuation data in each pair of different flight phases, resulting in positively correlated combination phase data, negatively correlated combination phase data, and uncorrelated combination phase data.

[0011] A preset threshold for predicting abnormal fluctuations in data is used to mark the data in the positive correlation combination stage that exceeds the threshold for predicting abnormal fluctuations, thereby obtaining positive correlation abnormal fluctuation marked stage data, where the positive correlation abnormal fluctuation marked stage data refers to the relative fluctuation amplitude ratio.

[0012] The negative correlation combined stage data is marked to exceed the data prediction abnormal fluctuation threshold, and negative correlation abnormal fluctuation marked stage data is obtained. The negative correlation abnormal fluctuation marked stage data refers to the relative fluctuation amplitude ratio.

[0013] The unrelated combination stage data is marked to exceed the data prediction abnormal fluctuation threshold, resulting in unrelated abnormal wave marked stage data, where the unrelated abnormal wave marked stage data refers to the relative fluctuation amplitude ratio;

[0014] The non-overlapping isotropic phase data among the positively correlated isotropic phase data, negatively correlated isotropic phase data, and uncorrelated isotropic phase data are removed to obtain the valid flight data to be tested after removal.

[0015] The valid flight data to be tested is subjected to time alignment, sampling frequency unification, and data format standardization to obtain preprocessed flight data.

[0016] Preferably, based on the preprocessed flight data, statistical analysis of health feature differences is performed to obtain candidate health features to be tested;

[0017] The health monitoring parameters to which the candidate health features to be tested belong are classified into hierarchical indicators from level one to level five to obtain multi-level structured health indicators. The hierarchical indicators from level one to level five include bottom-level sensor data, component operating parameters, subsystem status indicators, subsystem comprehensive indicators, and overall health status.

[0018] The multi-level structured health indicators are integrated layer by layer from the underlying sensor data to construct a multi-level feature parameter spectrum;

[0019] Based on the multi-level characteristic parameter spectrum, a hierarchical health status assessment is performed to obtain the overall system health status assessment result.

[0020] Preferably, based on the overall system health status assessment results and multi-level feature parameter spectra, a normal state baseline model is constructed for the operating status of each system to obtain an unsupervised health baseline model;

[0021] Extract single-dimensional parameters, multi-parameter comprehensive indicators, and time-series trend parameters that change over time from the multi-level feature parameter spectrum;

[0022] Based on the unsupervised health baseline model, anomaly detection is performed on single-dimensional parameters and multi-parameter comprehensive indicators to obtain local anomaly detection results and distribution anomaly detection results.

[0023] Based on the unsupervised health baseline model, the performance degradation trend of time-series trend parameters is predicted, and the trend prediction results are obtained.

[0024] Based on the local anomaly detection results, distribution anomaly detection results, and trend prediction results, the system health status is assessed to determine the early warning level, resulting in graded fault early warning information.

[0025] Preferably, based on the graded fault warning information, the structural manual, wiring manual and troubleshooting manual information of the aircraft are obtained, and based on the structural manual, wiring manual and troubleshooting manual information, the input-output relationship and control dependency relationship of the airborne system components are extracted to obtain the basic structural feature information of the system.

[0026] Based on the basic structural feature information of the system, a basic graph structure is constructed for the fault propagation logic to obtain an initial fault propagation feature graph.

[0027] Acquire airborne multi-source information, associate the airborne multi-source information with the initial fault transmission feature map, and obtain standardized system observation evidence nodes;

[0028] Based on the initial fault propagation feature map and standardized system observation evidence nodes, the fault propagation path is correlated and fused to obtain a directed graph of system failure state.

[0029] Preferably, fault diagnosis constraints are constructed based on the mapping relationship between the standardized system observation evidence nodes and corresponding nodes in the directed graph of the system failure state.

[0030] Based on the aforementioned fault diagnosis constraints and the directed graph of system failure states, statistical analysis is performed on all component failure combinations that lead to system anomalies to obtain component failure combinations.

[0031] The failure combinations of the components are judged to be contradictory and conflicting, and mutually contradictory failure combinations are filtered out and a set of failure conflicts is output.

[0032] Statistical analysis of the correlation and impact relationships of component failure combinations is performed to obtain the logical correlation relationships of failure combinations. The overall logical rationality of the logical correlation relationships of failure combinations is verified, and the logical verification results are output.

[0033] Preferably, based on the logical verification results, the minimum conflict set is identified and the minimum hit set is reversed from the fault conflict set to obtain the fault root cause prediction result.

[0034] Based on the predicted root causes of the faults, the causes of the faults are prioritized and their propagation paths are traced to obtain a list of root causes and a causal chain of fault propagation.

[0035] Based on airborne multi-source information, a list of root causes of failures and a causal chain of failure propagation, the operational status of each related system of the aircraft is compared with anomalies to obtain cascaded anomaly correlation information between systems.

[0036] Based on the inter-system cascaded anomaly correlation information, the source of cross-system coupled faults is located, and the fault location result is obtained.

[0037] Compared with the prior art, the present invention has the following characteristics and beneficial effects:

[0038] By segmenting flight phases, classifying different flight waves based on their correlations, removing invalid data, and standardizing the process, we effectively avoid interference from different flight conditions, eliminate random and redundant data, and retain the characteristics of valid flight data under test. This significantly reduces the interference of invalid data on subsequent health assessments and fault detection, improving the reliability of basic data. We employ unsupervised health baseline modeling, eliminating reliance on historical fault samples and adapting to the practical application scenarios where civil aviation and military aircraft fault samples are scarce. Combining static anomaly detection with dynamic degradation trend prediction, we can identify both sudden, obvious faults and predict slow performance degradation risks in advance, achieving dual capabilities of post-fault alarm and pre-fault trend warning. Through a series of scientific analyses (such as determining the constraint relationships between faults, screening contradictory faults, checking the rationality of logic, and analyzing abnormal correlations between different systems), we can accurately identify which fault combinations are contradictory and precisely pinpoint the true source of the fault. Especially for faults that are difficult to detect using traditional methods and involve the interaction of multiple systems, we can accurately locate them, greatly shortening troubleshooting time, reducing maintenance costs, and making aircraft maintenance more efficient and flight safer. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating the steps of an aircraft fault early warning method, which is the main feature of this embodiment. Detailed Implementation

[0040] The present invention will be further described in detail below with reference to the following embodiments.

[0041] Reference Figure 1 An aircraft malfunction early warning method, the method comprising the following steps:

[0042] Step S1: Obtain the aircraft's raw flight data. Based on the raw flight data, perform multi-level health indicator classification, bottom-up data fusion and summarization, and hierarchical health status assessment to obtain a multi-level characteristic parameter spectrum and the overall system health status assessment results.

[0043] Step S2: Based on the multi-level characteristic parameter spectrum and the overall system health status assessment results, determine the warning level of the aircraft's system health status, obtain graded fault warning information, acquire and construct the basic graph structure of the fault transmission logic according to the aircraft's structural manual, wiring manual and troubleshooting manual, obtain the initial fault transmission feature graph, acquire airborne multi-source information, and based on the airborne multi-source information, the initial fault transmission feature graph and graded fault warning information, correlate and fuse the fault propagation path to obtain the directed graph of system failure state.

[0044] Step S3: Node mapping and diagnostic constraint construction are performed on the directed graph of system failure state to obtain fault diagnosis constraints. Based on the fault diagnosis constraints and the directed graph of system failure state, the causes of failure are prioritized and propagation paths are traced to obtain a list of root causes and a fault propagation causal chain. Based on the list of root causes and the fault propagation causal chain, the source of cross-system coupled faults is quickly located to obtain the fault location result.

[0045] Specifically, by segmenting flight phases, classifying different flight waves, removing invalid data, and standardizing the process, interference from different flight conditions is effectively avoided, random and redundant data is eliminated, and valid flight data characteristics are retained. This significantly reduces the interference of invalid data on subsequent health assessments and fault detection, improving the reliability of basic data. Unsupervised health baseline modeling is adopted, eliminating reliance on historical fault samples and adapting to the practical application scenarios where civil aviation and military aircraft fault samples are scarce. Combining static anomaly detection with dynamic degradation trend prediction, it can identify both sudden, obvious faults and predict slow performance degradation risks in advance, achieving dual capabilities of post-fault alarm and pre-fault trend warning. Through a series of scientific analyses (such as determining the constraint relationships between faults, screening contradictory faults, checking the rationality of logic, and analyzing abnormal correlations between different systems), it can accurately identify which fault combinations are contradictory, precisely pinpointing the true source of the fault. Especially for faults that are difficult to detect using traditional methods and involve the interaction of multiple systems, we can accurately locate them, greatly shortening troubleshooting time, reducing maintenance costs, and making aircraft maintenance more efficient and flight safer.

[0046] The specific step S1 includes the following sub-steps:

[0047] The raw flight data is divided into four phases: takeoff and climb, cruise and steady-state, descent and approach, and landing and taxiing, resulting in takeoff phase data, cruise phase data, descent phase data, and landing phase data.

[0048] The abnormal fluctuation data between different flight phases are classified and statistically analyzed according to the data of takeoff, cruise, descent and landing, resulting in positively correlated combination data, negatively correlated combination data and no-correlation combination data.

[0049] A threshold for predicting abnormal fluctuations in data is preset. Data in the positive correlation combination stage that exceeds the threshold for predicting abnormal fluctuations is marked to obtain data in the positive correlation abnormal fluctuation marked stage. The data in the positive correlation abnormal fluctuation marked stage refers to the ratio of relative fluctuation amplitude.

[0050] The negative correlation combined stage data is marked to exceed the data prediction abnormal fluctuation threshold, and the negative correlation abnormal fluctuation marked stage data refers to the relative fluctuation amplitude ratio.

[0051] The data in the uncorrelated combination stage are marked to exceed the threshold of abnormal fluctuations predicted by the data, and the uncorrelated abnormal fluctuation marked stage data refers to the relative fluctuation amplitude ratio.

[0052] By removing non-overlapping outside wave stage data from the positively correlated outside wave stage data, the negatively correlated outside wave stage data, and the uncorrelated outside wave stage data, the valid flight data to be tested after removal is obtained.

[0053] The valid flight data to be tested is time-aligned, sampled at the same frequency, and formatted to obtain preprocessed flight data.

[0054] Based on the preprocessed flight data, statistical analysis of health characteristic differences is performed to obtain candidate health characteristics to be tested.

[0055] The health monitoring parameters of the candidate health features to be tested are classified into hierarchical indicators from level one to level five to obtain multi-level structured health indicators. The hierarchical indicators from level one to level five include bottom-level sensor data, component operating parameters, subsystem status indicators, subsystem comprehensive indicators, and overall health status.

[0056] By integrating the underlying sensor data layer by layer, a multi-level feature parameter spectrum is constructed from the multi-level structured health indicators.

[0057] Based on the multi-level characteristic parameter spectrum, a hierarchical health status assessment is performed to obtain the overall system health status assessment results.

[0058] Specifically, for example, the C919 aircraft's rapid access recorder collects raw flight data (engine speed, oil pressure, temperature, vibration, fuel flow, control surface deflection, atmospheric parameters, etc.), takeoff and climb phase data: data from takeoff taxiing to the entire process of climbing to cruise altitude (e.g., 0-15 minutes, covering takeoff taxiing (speed increasing from 0 to 150 knots), wheel lift-off, and climbing to a cruise altitude of 9000 meters, focusing on engine speed (increasing from 20% idle speed to 85% cruise speed), takeoff flap deflection (0°→15°), landing gear retraction and extension status, etc.), and cruise stability phase data: high-altitude level flight and stable flight data (e.g., For example, from minutes 16 to 90, the aircraft was flying at an altitude of 9000 meters, maintaining a stable speed of around 450 knots with minimal data fluctuations. Key data included engine high-pressure rotor speed (stable at 85% ± 2%), fuel flow rate (2200 kg / h ± 50 kg / h), and cabin temperature (24℃ ± 1℃). Descent and approach phase data: data from the start of descent to runway alignment (e.g., from minutes 91 to 110, descending from 9000 meters to 1000 meters and aligning with the runway, key data included engine speed (decreasing from 85% to 60%) and descent rate (2000 ft / min ± 300 ft / min). Landing flap deflection angle (15°→30°, etc.), landing taxiing phase data: touchdown to runway taxiing and parking data (e.g., 111-120 minutes, touchdown (vertical load reaches 1.8G), runway taxiing to parking position, focusing on brake pressure (0→18MPa→0), landing gear load, taxiing speed, etc.). Existing technologies mostly use uniform processing throughout or a rough two-stage segmentation (takeoff and landing or cruise), which easily leads to confusion of characteristic data in different stages (e.g., high-speed data in the takeoff stage is mixed with stable speed data in the cruise stage), making it impossible to accurately capture the parameter characteristics of each stage. This step uses a four-stage precise segmentation to clarify the scene attributes of the data in each stage, which is crucial for subsequent abnormal fluctuations. Correlation analysis lays the foundation by pairing the four phases of takeoff, cruise, descent, and landing (e.g., takeoff-cruise, takeoff-descent, takeoff-landing, cruise-descent, cruise-landing, descent-landing). The correlation between these fluctuations is statistically analyzed based on the fluctuations of core parameters (engine temperature, vibration, fuel flow, pipeline oil pressure, etc.). Positive correlation was found in the combined phase data: selecting engine high-pressure turbine temperature and engine vibration acceleration as two core parameters, the analysis revealed that in the "takeoff-cruise" combination, when the engine temperature increased from 650℃ to 820℃ (fluctuation of 26%), the vibration acceleration increased from 0.8g to 1.2g (fluctuation of 50%), showing a clear trend of simultaneous increase and decrease (correlation coefficient 0.87), indicating a positive correlation.Meanwhile, in the "cruise-descent" combination, when the fuel flow rate decreased from 2200 kg / h to 1500 kg / h (fluctuation of 32%), the engine speed decreased from 85% to 60% (fluctuation of 30%), both decreasing synchronously (correlation coefficient 0.91), which is also classified as data in the positive correlation combination phase. For the negative correlation combination phase data: selecting fuel flow rate and fuel line pressure as two parameters, in the "takeoff-cruise" combination, when the fuel flow rate increased from 1800 kg / h to 2200 kg / h (fluctuation of 22%), the fuel line pressure decreased from 12 MPa to 10.5 MPa (fluctuation of 12.5%), showing an inverse correlation (correlation coefficient -0.76); in the "descent-landing" combination, the brake pressure increased from 0 to 18 MPa... When fluctuation is unlimited (a), the taxiing speed drops from 120 knots to 0 (fluctuation 100%), and the two fluctuate in opposite directions (correlation coefficient -0.93), which is classified as negatively correlated combination stage data. For uncorrelated combination stage data: selecting cabin temperature and engine speed as two parameters, statistics are compiled for all combination stages. The cabin temperature remains stable at 23-25℃ (fluctuation ≤8%), while the engine speed fluctuates significantly in each stage (65% fluctuation during takeoff, 4% during cruise, and 32% during descent), with no correlation between the two fluctuations (correlation coefficient 0.08). Simultaneously, the landing gear retraction / extension status and cabin humidity fluctuations are also uncorrelated (correlation coefficient 0.12), classified as uncorrelated combination stage data. Existing technology only analyzes anomalies in parameters within a single stage. The previous detection method ignored the correlation between parameter fluctuations in different flight phases, which easily led to misjudgments of anomalies in a single phase (e.g., a small fluctuation in a parameter in a certain phase, if not combined with the correlation with other phases, is easily misjudged as an anomaly). This step, through pairwise phase correlation statistics, explores the linkage patterns of parameters between different phases, classifies the data according to correlation type, and solves the shortcomings of the existing technology in isolated analysis of a single phase and ignoring the correlation between phases. This provides a more comprehensive judgment basis for subsequent anomaly labeling, reduces the probability of misjudgment, and positively correlates the anomaly labeling phase data: taking engine temperature and vibration acceleration in the "takeoff-cruise" combination as an example, the historical average relative fluctuation ratios are 8% and 10%, respectively. During this flight, the engine temperature fluctuation ratio reached 18% at a certain moment (exceeding 15%, an anomaly). Fluctuation amplitude thresholds (e.g., relative fluctuations of parameters exceeding 15% (obtained through historical experience data statistics) are judged as suspicious interference anomalies). The corresponding vibration acceleration fluctuation ratio reaches 22% (exceeding 15%), indicating synchronous anomalies. This is marked as positively correlated anomaly marker data, and the relative fluctuation ratios of the two are recorded (18%, 22%). Negatively correlated anomaly marker data: Taking brake pressure and taxi speed in the "descent-landing" combination as an example, the historical average relative fluctuation ratios are 12% and 14%, respectively. During this flight, at a certain moment, the brake pressure fluctuation ratio reached 19%, and the corresponding taxi speed fluctuation ratio reached 17%. The two are reversed anomalies, marked as negatively correlated anomaly marker data, and the relative fluctuation ratios are recorded (19%, 17%).Unrelated anomaly marker phase data: Taking cabin temperature as an example, the historical average relative fluctuation ratio is 5%. During this flight, at a certain moment, the cabin temperature experienced a brief fluctuation due to the air conditioning system, resulting in a relative fluctuation ratio of 16% (exceeding 15%). However, the corresponding engine speed fluctuation ratio was only 3% (normal). Since the two are unrelated, this data is marked as unrelated anomaly marker phase data, and the relative fluctuation ratio is recorded (16%, relative fluctuation amplitude ratio = (current fluctuation value - historical average fluctuation value) / historical average fluctuation value × 100%). Invalid anomaly data: For example, at the moment of landing (115 minutes), the engine exhaust temperature experienced a single small fluctuation with a relative fluctuation ratio of 16%, which is marked as unrelated anomaly data, but... The fluctuation only occurred in the "descent-landing" combination and did not overlap with the anomaly data of the other five phase combinations (such as takeoff-cruise, cruise-descent, etc.). Furthermore, the fluctuation lasted only 1 second (single sampling), indicating it was an occasional, transient interference and was therefore deemed invalid and discarded. Valid flight data to be tested: If the retained anomaly data all exhibited anomalies with "multi-phase combination overlap," for example, an anomaly with an engine temperature fluctuation ratio of 18% appearing in both the "takeoff-cruise" and "cruise-descent" combinations, and the corresponding vibration acceleration anomaly appeared synchronously, this was considered a persistent and correlated anomaly and was therefore deemed valid. Finally, after discarding invalid data, valid flight data to be tested was obtained. By removing non-overlapping anomaly data and selecting persistent and correlated valid anomaly data, the problems of invalid anomaly data interference and data redundancy in existing technologies were solved, reducing subsequent data processing volume and improving the accuracy of fault warnings, while avoiding false warnings caused by transient interference.The valid flight data to be tested is processed by: aligning the time axis, standardizing the sampling frequency to 1Hz, and standardizing the data units / format, to finally obtain preprocessed flight data. By comparing preprocessed flight data with historical normal fleet data, parameters with discrepancies are screened out to obtain candidate health characteristics to be tested (such as engine high-pressure rotor speed deviation, hydraulic system oil pressure fluctuation deviation, etc.). Level 1 (bottom-level sensor data): raw values ​​from temperature, pressure, and vibration sensors; Level 2 (component operating parameters): engine rotor speed, hydraulic pump output pressure, and fuel pump flow; Level 3 (subsystem status indicators): engine subsystem health and hydraulic subsystem stability indicators; Level 4 (subsystem comprehensive indicators): comprehensive scores of the power subsystem, flight control subsystem, and environmental control subsystem; Level 5 (overall aircraft health status): the overall flight safety and health level of the entire aircraft. Starting from the bottom-level sensor data, data is aggregated and integrated layer by layer upwards to form a complete characteristic parameter spectrum (multi-level characteristic parameter spectrum) from sensor → component → subsystem → subsystem → whole aircraft. For example, bottom-level integration (Level 1 → Level 2): ​​the raw data from engine temperature, vibration, and speed sensors (Level 1) are integrated and calculated to obtain component operating parameters such as engine high-pressure rotor speed, turbine temperature fluctuation, and vibration acceleration. Level 2): ​​Clearly define the calculation logic of "sensor raw data → component parameters" (e.g., speed = sensor acquired value / rated speed × 100%); Subsystem integration (Level 2 → Level 3): Calculate the engine subsystem health (Level 3) by using a weighted summation method (weights of 40%, 30%, and 30% respectively) for component parameters such as speed, temperature, and vibration within the engine subsystem; similarly, integrate component parameters such as hydraulic pumps and steering gear to obtain the hydraulic subsystem stability index; Subsystem integration (Level 3 → Level 4): Integrate the parameters within the power subsystem... The status indicators of the engine subsystem, fuel subsystem, etc. (Level 3) are weighted (60% for engine subsystem and 40% for fuel subsystem) to calculate the comprehensive score of the power subsystem (Level 4). Similarly, the indicators of the hydraulic, flight control, and environmental control subsystems are integrated to obtain the comprehensive scores of the corresponding subsystems. Overall system integration (Level 4 to Level 5): The comprehensive scores of the four subsystems of power, hydraulic, flight control, and environmental control (Level 4) are weighted according to preset weights (40% for power, 25% for hydraulic, 25% for flight control, and 10% for environmental control) to calculate the overall health level of the system (Level 5). The system health status is evaluated by scoring each layer according to the multi-level characteristic parameter spectrum, and the final output is the overall system health status evaluation result (for example, Level 1 evaluation (bottom sensor data): all raw sensor data are within the normal range (such as temperature 780-830℃, pressure 10-12MPa), with no abnormalities, and a score of 100 points; Level 2 evaluation (component operating parameters): the hydraulic pump output pressure fluctuation (±0.8MPa) exceeds the normal range (±0.5MPa), deducting 10 points; the servo motor response speed (0.6° / s) exceeds the normal range (0.5° / s±0.1° / s), deducting 5 points; the parameters of other components are normal, and the Level 2 comprehensive score is 85 points).Level 3 Assessment (Subsystem Status Indicators): Hydraulic subsystem stability index (fluctuation coefficient 0.07) exceeds the normal range (≤0.05), deduct 15 points; Engine subsystem health (88 points) is close to the normal threshold (90 points), deduct 2 points; Flight control servo subsystem response index (0.25s) exceeds the normal range (≤0.2s), deduct 8 points; Level 3 overall score: 75 points; Level 4 Assessment (Subsystem Comprehensive Indicators): Power subsystem (86 points), Hydraulic subsystem (82 points), Flight control subsystem... The overall system score (89 points) is below the normal threshold (90 points), resulting in deductions of 4, 8, and 1 points respectively; the environmental control subsystem (95 points) is normal, with no deductions; the fourth-level comprehensive score is 86 points; the fifth-level assessment (overall machine health status): combining the scores and weights of each subsystem at the fourth level, the overall machine health score is calculated to be 86.2 points, corresponding to level two. The final output of the overall machine system health status assessment result is: slight abnormality in the hydraulic subsystem, slight abnormality in the power subsystem, and slight abnormality in the flight control subsystem, resulting in a level two warning for the overall machine health status.

[0059] The specific step S2 includes the following sub-steps:

[0060] Based on the overall system health status assessment results and multi-level characteristic parameter spectra, a normal state baseline model is constructed for the operating status of each system, resulting in an unsupervised health baseline model.

[0061] Single-dimensional parameters, multi-parameter comprehensive indicators, and time-series trend parameters that change over time are extracted from the multi-level feature parameter spectrum.

[0062] Based on the unsupervised health baseline model, anomaly detection was performed on single-dimensional parameters and multi-parameter comprehensive indicators to obtain local anomaly detection results and distribution anomaly detection results.

[0063] Based on the unsupervised health baseline model, the performance degradation trend of time-series trend parameters is predicted, and the trend prediction results are obtained.

[0064] Based on the results of local anomaly detection, distribution anomaly detection, and trend prediction, the system health status is assessed to determine the early warning level, resulting in graded fault early warning information.

[0065] Based on the graded fault warning information, obtain the structural manual, wiring manual and troubleshooting manual information of the aircraft. Based on the structural manual, wiring manual and troubleshooting manual information, extract the input-output relationship and control dependency relationship of the airborne system components to obtain the basic structural characteristic information of the system.

[0066] Based on the basic structural characteristics of the system, a basic graph structure is constructed for the fault propagation logic to obtain an initial fault propagation characteristic graph.

[0067] By acquiring airborne multi-source information and correlating it with the initial fault propagation feature map, standardized system observation evidence nodes are obtained.

[0068] Based on the initial fault propagation feature map and standardized system observation evidence nodes, the fault propagation path is correlated and fused to obtain a directed graph of system failure state.

[0069] Specifically, based on the overall aircraft health assessment results and multi-level characteristic spectra, historical normal flight data is used to establish normal state baseline models (unsupervised health baseline models) for each system, including the engine, hydraulic, flight control, and environmental control systems. This is equivalent to defining a normal fluctuation range for each parameter. For example, the engine system baseline model has the following normal fluctuation ranges: high-pressure rotor speed 85% ± 3%, high-pressure turbine temperature 800 ± 20℃, and vibration acceleration 0.8 ± 0.3g. The hydraulic system baseline model has the following normal fluctuation ranges: hydraulic pump output pressure 10.8 ± 0.5 MPa, hydraulic pipeline pressure 11 ± 0.6 MPa, and servo hydraulic supply pressure 10.5 ± 0.7 MPa. The flight control system baseline model has the following normal fluctuation ranges: control surface fine-tuning angle 0.5 ± 0.2°, and servo response delay ≤ 0.2s. The overall aircraft baseline model has the following health scores: overall aircraft health score ≥ 90 points, comprehensive scores for each subsystem ≥ 90 points, and no single parameter exceeding the normal fluctuation range. The unsupervised health baseline model can be automatically updated and optimized based on subsequent new normal flight data, ensuring the accuracy and adaptability of the baseline without the need for frequent manual adjustments.Single-dimensional parameters: such as temperature from a single sensor, or oil pressure value from a single sensor; multi-parameter comprehensive indicators: such as combined indicators of engine speed, temperature, and vibration; time-series trend parameters: such as the time-series change curve of slow engine wear and deterioration over a long period; single-parameter baseline comparison: detecting local anomalies (such as oil pressure exceeding the standard at a single point at a certain moment); multi-parameter comprehensive indicator baseline distribution comparison: detecting distribution anomalies (multiple parameters deviating from the normal range); time-series trend parameter prediction: predicting performance degradation trends (over the next 50 years). (The engine will continue to deteriorate during flight hours). For example, the results of local anomaly detection (single-dimensional parameter): comparing the single parameter with the normal fluctuation range of the baseline model, two local anomalies were detected: the hydraulic pump output pressure (10.2MPa) was lower than the baseline lower limit (10.3MPa), and the control surface fine adjustment angle (0.8°) exceeded the baseline upper limit (0.7°); the remaining single parameters were all within the normal range, with no local anomalies; the results of distribution anomaly detection (multi-parameter comprehensive index): comparing the multi-parameter comprehensive index with the normal distribution range of the baseline model, one distribution anomaly was detected: the hydraulic system pressure + flow rate joint index (1.8) exceeded the normal threshold (1.2), indicating that the hydraulic system... Multiple parameters deviated from the normal state, which is a system-level distribution anomaly; the engine joint index (2.1) exceeded the normal threshold (1.5), which is a slight distribution anomaly; trend prediction results (time-series trend parameters): using the LSTM time-series prediction algorithm, the performance degradation trend of the time-series trend parameters in the next 50 flight hours was predicted, and two trend anomalies were obtained: the hydraulic pump output pressure will continue to decrease, and it is expected to drop to 9.8MPa after 50 flight hours (below the baseline lower limit of 10.3MPa); the engine high-pressure rotor speed will slowly decrease, and it is expected to drop to 83% after 50 flight hours (close to the baseline lower limit of 82%); the other time-series parameters did not show obvious degradation trends. If the criteria for determining the warning level are: 10 points deducted for each local anomaly, 20 points deducted for each distribution anomaly, and 25 points deducted for each trend anomaly; 100-90 points is a Level 1 minor warning, 89-80 points is a Level 2 attention warning, 79-70 points is a Level 3 fault warning, and <70 points is a Level 4 emergency alarm, then the judgment process for this case is as follows: 2 local anomalies (deduct 20 points), 2 distribution anomalies (deduct 40 points), and 2 trend anomalies (deduct 50 points). The base score is 100 points, and 110 points are deducted (based on the minimum score of 70 points). Combined with the overall aircraft health assessment results (Level 2), the overall judgment is a Level 3 fault warning. The graded fault warning information is as follows: "Level 3 fault warning for the hydraulic subsystem, Level 2 attention warning for the power subsystem, and Level 1 minor warning for the flight control subsystem; abnormal hydraulic pump output pressure, abnormal control surface deflection angle, system-level distribution anomalies in the hydraulic system, and performance degradation trends in the hydraulic pump and engine. The hydraulic pump inspection and maintenance must be completed within 50 flight hours."Based on the aircraft structure manual, wiring manual, and troubleshooting manual, the input-output relationships (e.g., hydraulic pump output pressure to servos) and control dependencies (e.g., flight control computer controlling control surface actuators) of airborne components are analyzed to obtain the basic structural characteristics of the system. Based on these basic structural characteristics and combined with graded fault warning information, an initial fault propagation characteristic diagram is drawn. For example, fault propagation logic 1 (hydraulic system): internal wear of the hydraulic pump → hydraulic pump output pressure decreases (below baseline) → insufficient hydraulic line pressure → insufficient servo power supply → sluggish servo response → abnormal flight control surface deflection angle (beyond baseline); fault propagation logic 2 (related effects): hydraulic pump output pressure decreases → insufficient hydraulic pressure in the braking system → sluggish braking response during landing taxiing (no obvious abnormalities in this flight, but potential risks exist); initial fault propagation characteristic diagram: hydraulic pump (candidate fault source) → hydraulic lines → servos → control surfaces (fault terminal); hydraulic pump → Braking system (potential fault terminal); each node is labeled with corresponding abnormal parameters (e.g., the hydraulic pump node is labeled "output pressure 10.2MPa, baseline 10.3-11.3MPa"). Access the onboard multi-source information from this flight and bind it to each node of the initial fault propagation feature map to supplement the observational evidence of the nodes, forming standardized system observational evidence nodes: for example, onboard alarm information: the "low hydraulic system pressure" alarm (occurred 3 times, each lasting 2 seconds) during this flight is bound to the "hydraulic pipeline" node as observational evidence of insufficient pipeline pressure; real-time sensor data: hydraulic pump output pressure time-series data (continuously decaying, from 10.5MPa to 10.2MPa), servo response speed data (0.6° / s, exceeding the baseline), etc. Do not bind to the "hydraulic pump" or "servo motor" nodes as observational evidence of component anomalies; Airborne maintenance messages: In the past 10 flight hours of this aircraft, the hydraulic system maintenance message showed "minor leakage of hydraulic pump seals", bind to the "hydraulic pump" node as auxiliary observational evidence of hydraulic pump wear; Standardization processing: Standardize all observational evidence according to the format of "node name-observational evidence type-evidence content-confidence", for example: hydraulic pump node (observational evidence type: maintenance message + sensor data; evidence content: seal leakage, output pressure attenuation; confidence: 85%). By associating and fusing the initial fault propagation feature map with standardized system observation evidence nodes, the credibility and anomaly degree of fault propagation are supplemented, the fault propagation path is optimized, and finally a directed graph of system failure state with causal relationship and observation evidence is obtained (e.g., the complete fault propagation path of "internal wear of hydraulic pump (source, moderate anomaly) → hydraulic pump output pressure decay → insufficient hydraulic pipeline pressure → insufficient servo motor power → slow servo motor response → abnormal servo surface deflection angle (terminus, slight anomaly)" is obtained, and the observation evidence, credibility and anomaly degree of each node are marked, so that the step-by-step transmission process of the fault can be seen intuitively).

[0070] The specific step S3 includes the following sub-steps:

[0071] Based on the standardized mapping relationship between system observation evidence nodes and corresponding nodes in the directed graph of system failure states, fault diagnosis constraints are constructed.

[0072] Based on the fault diagnosis constraints and the directed graph of system failure states, the failure combinations of all components that cause system abnormalities are statistically analyzed to obtain component failure combinations.

[0073] The system determines the contradictions and conflicts among component failure combinations, filters out mutually contradictory failure combinations, and outputs a set of failure conflicts.

[0074] Statistical analysis of the correlation and impact relationships of component failure combinations is performed to obtain the logical correlation relationships of failure combinations. The overall logical rationality of the logical correlation relationships of failure combinations is verified, and the logical verification results are output.

[0075] Based on the logical verification results, the minimum conflict set is identified and the minimum hit set is deduced from the fault conflict set to obtain the fault root cause prediction result.

[0076] Based on the predicted root causes of the faults, the causes of the faults are prioritized and their propagation paths are traced to obtain a list of root causes and a causal chain of fault propagation.

[0077] Based on airborne multi-source information, a list of root causes of failures, and a causal chain of failure propagation, the operational status of each related system of the aircraft is compared with anomalies to obtain cascaded anomaly correlation information between systems.

[0078] Based on the cascading anomaly correlation information between systems, the source of cross-system coupled faults is located, and the fault location results are obtained.

[0079] Specifically, constraints are set based on the mapping relationship between "evidence nodes - component nodes" in the directed graph of system failure states. For example, logical constraint 1: the observation evidence of the "abnormal control surface deflection angle" node (slow servo response) must be associated with "insufficient hydraulic system pressure" or "flight control actuator failure," and at least one of them must be true (if neither is true, then the abnormal control surface deflection angle has no reasonable cause and the constraint is violated); logical constraint 2: "insufficient hydraulic line pressure" must be associated with "hydraulic pump output pressure decay" or "hydraulic line leakage," and at least one of them must be true; quantitative constraint: the abnormality degree (moderate) of "hydraulic pump output pressure decay," the corresponding fault cause (such as wear, seal leakage), and its impact range must cover "insufficient hydraulic line pressure" and "insufficient servo power," otherwise the constraint is violated; constraint priority: logical constraints are higher than quantitative constraints. If a logical constraint is violated, the fault combination is directly determined to be invalid. Candidate failure combinations (e.g., combination 1: internal wear of hydraulic pump; combination 2: flight control actuator failure; combination 3: hydraulic line leakage; combination 4: internal wear of hydraulic pump + hydraulic line leakage; combination 5: internal wear of hydraulic pump + flight control actuator failure; flight control actuator failure + hydraulic line leakage), combined with fault diagnosis constraints, combinations that violate constraints are eliminated (combination 6: flight control actuator failure + hydraulic line leakage, which cannot explain the observational evidence of abnormal control surface deflection angle, violating logic constraint 1; combination 3: hydraulic line leakage, which is not supported by observational evidence, violating quantification constraint), so the component failure combinations are combination 1, combination 2, combination 4, and combination 5.Fault Conflict Sets: For example, combination 1 (internal wear of hydraulic pump) and combination 2 (flight control actuator failure): Observational evidence shows "low hydraulic system pressure," but combination 2 (flight control actuator failure) cannot explain this evidence, and both are single fault causes, so they cannot be simultaneously established (if they were simultaneously established, the observational evidence should show both hydraulic and flight control abnormalities; in this case, only the hydraulic abnormality is obvious), thus it is determined to be a conflicting combination; combination 4 (internal wear of hydraulic pump + hydraulic pipeline leakage) and combination 2 (flight control actuator failure): Similarly, combination 2 cannot explain the evidence of low hydraulic pressure, conflicting with combination 4, thus it is determined to be a conflicting combination; Statistically analyze the mutual influence relationship of each failure combination to verify the logical coherence: For example, combination 1 (internal wear of hydraulic pump): The logical relationship is "hydraulic pump wear → pressure decay → insufficient pipeline pressure → insufficient servo power → abnormal control surface deflection," which is consistent with the transmission path of the directed graph of system failure states, and all observational evidence (low hydraulic pressure)... The combination of faults (such as abnormal control surface deflection and hydraulic pump seal leakage) can be supported, the logic is sound, and the verification is qualified (this is just an example, and the same applies to other faults). Finally, the logic verification result is output. Based on the logic verification result, the most likely and least number of fault combinations are selected from the fault conflict set, and the root cause prediction result is deduced in reverse (for example, the initial identification is likely a hydraulic pump wear failure). The predicted faults are sorted by probability priority, and the fault propagation link is traced: internal wear of hydraulic pump → oil pressure decline → insufficient power of servo motor → abnormal cruise control surface fine adjustment, forming a fault root cause list + fault propagation causal chain. Combined with airborne multi-source data, the abnormal linkage between the power system, hydraulic system, and flight control system is compared to obtain the cascaded abnormal correlation information between the systems, and finally the precise location is obtained: for example, the abnormality of this aircraft is not a fault of the flight control itself, but the cross-system coupling source of hydraulic system pump wear. The final fault location result is output, which directly guides the maintenance personnel to troubleshoot the fault.

[0080] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for early warning of aircraft malfunctions, characterized in that, Includes the following steps: Step S1: Obtain the aircraft's raw flight data, perform multi-level health indicator classification, bottom-up data fusion and summarization, and hierarchical health status assessment based on the raw flight data, and comprehensively obtain the multi-level characteristic parameter spectrum and the whole system health status assessment results. Step S2: Based on the multi-level feature parameter spectrum and the whole system health status assessment results, determine the warning level of the aircraft's system health status, obtain graded fault warning information, acquire and construct the basic graph structure of the fault transmission logic according to the structural manual, wiring manual and troubleshooting manual of the aircraft, obtain the initial fault transmission feature graph, acquire airborne multi-source information, and perform correlation and fusion of the fault propagation path according to the airborne multi-source information, the initial fault transmission feature graph and the graded fault warning information to obtain the directed graph of system failure status; Step S3: Node mapping and diagnostic constraint construction are performed on the directed graph of the system failure state to obtain fault diagnosis constraints. Based on the fault diagnosis constraints and the directed graph of the system failure state, the causes of the fault are prioritized and the propagation path is traced to obtain a list of root causes and a fault propagation causal chain. Based on the list of root causes and the fault propagation causal chain, the source of the cross-system coupled fault is quickly located to obtain the fault location result.

2. The aircraft fault early warning method according to claim 1, characterized in that, Step S1 includes: The raw flight data is divided into four phases: takeoff and climb, cruise and steady-state, descent and approach, and landing and taxiing, resulting in takeoff phase data, cruise phase data, descent phase data, and landing phase data. The takeoff, cruise, descent, and landing data are classified and statistically analyzed to determine the correlation between abnormal fluctuation data in each pair of different flight phases, resulting in positively correlated combination phase data, negatively correlated combination phase data, and uncorrelated combination phase data. A preset threshold for predicting abnormal fluctuations in data is used to mark the data in the positive correlation combination stage that exceeds the threshold for predicting abnormal fluctuations, thereby obtaining positive correlation abnormal fluctuation marked stage data, where the positive correlation abnormal fluctuation marked stage data refers to the relative fluctuation amplitude ratio. The negative correlation combined stage data is marked to exceed the data prediction abnormal fluctuation threshold, and negative correlation abnormal fluctuation marked stage data is obtained. The negative correlation abnormal fluctuation marked stage data refers to the relative fluctuation amplitude ratio. The unrelated combination stage data is marked to exceed the data prediction abnormal fluctuation threshold, resulting in unrelated abnormal wave marked stage data, where the unrelated abnormal wave marked stage data refers to the relative fluctuation amplitude ratio; The non-overlapping isotropic phase data among the positively correlated isotropic phase data, negatively correlated isotropic phase data, and uncorrelated isotropic phase data are removed to obtain the valid flight data to be tested after removal. The valid flight data to be tested is subjected to time alignment, sampling frequency unification, and data format standardization to obtain preprocessed flight data.

3. The aircraft fault early warning method according to claim 2, characterized in that, Step S1 also includes: Based on the preprocessed flight data, statistical analysis of health feature differences is performed to obtain candidate health features to be tested; The health monitoring parameters to which the candidate health features to be tested belong are classified into hierarchical indicators from level one to level five to obtain multi-level structured health indicators. The hierarchical indicators from level one to level five include bottom-level sensor data, component operating parameters, subsystem status indicators, subsystem comprehensive indicators, and overall health status. The multi-level structured health indicators are integrated layer by layer from the underlying sensor data to construct a multi-level feature parameter spectrum; Based on the multi-level characteristic parameter spectrum, a hierarchical health status assessment is performed to obtain the overall system health status assessment result.

4. The aircraft fault early warning method according to claim 3, characterized in that, Step S2 includes: Based on the overall system health status assessment results and multi-level feature parameter spectra, a normal state baseline model is constructed for the operating status of each system to obtain an unsupervised health baseline model. Extract single-dimensional parameters, multi-parameter comprehensive indicators, and time-series trend parameters that change over time from the multi-level feature parameter spectrum; Based on the unsupervised health baseline model, anomaly detection is performed on single-dimensional parameters and multi-parameter comprehensive indicators to obtain local anomaly detection results and distribution anomaly detection results. Based on the unsupervised health baseline model, the performance degradation trend of time-series trend parameters is predicted, and the trend prediction results are obtained. Based on the local anomaly detection results, distribution anomaly detection results, and trend prediction results, the system health status is assessed to determine the early warning level, resulting in graded fault early warning information.

5. The aircraft fault early warning method according to claim 4, characterized in that, Step S2 also includes: Based on the graded fault warning information, obtain the structural manual, wiring manual and troubleshooting manual information of the aircraft. Based on the structural manual, wiring manual and troubleshooting manual information, extract the input-output relationship and control dependency relationship of the airborne system components to obtain the basic structural feature information of the system. Based on the basic structural feature information of the system, a basic graph structure is constructed for the fault propagation logic to obtain an initial fault propagation feature graph. Acquire airborne multi-source information, associate the airborne multi-source information with the initial fault transmission feature map, and obtain standardized system observation evidence nodes; Based on the initial fault propagation feature map and standardized system observation evidence nodes, the fault propagation path is correlated and fused to obtain a directed graph of system failure state.

6. The aircraft fault early warning method according to claim 5, characterized in that, Step S3 includes: Based on the mapping relationship between the standardized system observation evidence nodes and their corresponding nodes in the directed graph of the system failure state, fault diagnosis constraints are constructed. Based on the aforementioned fault diagnosis constraints and the directed graph of system failure states, statistical analysis is performed on all component failure combinations that lead to system anomalies to obtain component failure combinations. The failure combinations of the components are judged to be contradictory and conflicting, and mutually contradictory failure combinations are filtered out and a set of failure conflicts is output. Statistical analysis of the correlation and impact relationships of component failure combinations is performed to obtain the logical correlation relationships of failure combinations. The overall logical rationality of the logical correlation relationships of failure combinations is verified, and the logical verification results are output.

7. The aircraft fault early warning method according to claim 6, characterized in that, Step S3 also includes: Based on the logical verification results, the minimum conflict set identification and minimum hit set inverse deduction are performed on the fault conflict set to obtain the fault root cause prediction result. Based on the predicted root causes of the faults, the causes of the faults are prioritized and their propagation paths are traced to obtain a list of root causes and a causal chain of fault propagation. Based on airborne multi-source information, a list of root causes of failures and a causal chain of failure propagation, the operational status of each related system of the aircraft is compared with anomalies to obtain cascaded anomaly correlation information between systems. Based on the inter-system cascaded anomaly correlation information, the source of cross-system coupled faults is located, and the fault location result is obtained.

Images