Helicopter fleet operation safety state evaluation system based on multi-source data fusion

Abstract

The invention relates to the technical field of aircraft operation support and health management, in particular to a helicopter team operation safety state evaluation system based on multi-source data fusion. The system comprises a data alignment module, a residual calculation module and a closed-loop determination module. The system segments the flight parameters into action elements, and analyzes a maintenance log to generate a weight; the core of the method is that a theoretical baseline is called in response to an action type, and a situational health residual is calculated in combination with a weight; when the residual error exceeds a preset safety threshold value, the system generates a high-load task forbidding instruction and updates scheduling; according to the invention, by decoupling the intention of the driver and the mechanical response, the problem of high false alarm rate under violent maneuvering is solved, and the precise monitoring of the mechanical health state under the dynamic working condition is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of aircraft operation support and health management, in particular to a helicopter fleet operation safety state evaluation system based on multi-source data fusion. BACKGROUND

[0002] In the current helicopter fleet operation maintenance environment, the system needs to process the numerical flight parameters collected by the on-board sensor at a high frequency and the unstructured maintenance logs generated by the maintenance terminal. The existing safety evaluation scheme generally adopts a single physical modeling or fixed threshold comparison architecture, that is, it relies on physical model calculation theoretical value or directly monitors whether the parameter is out of limit. Although this scheme has basic monitoring capability in the steady cruise scene, due to the high coupling between flight parameters and pilot control intention, high false alarm rate often occurs due to sudden load change during sharp maneuvering. Moreover, there are essential differences in semantic expression and update frequency between maintenance data and flight data, which makes it difficult to effectively fuse multi-source heterogeneous data and form an information island. In addition, the traditional static threshold cannot adapt to the physical law during the running-in period after component replacement, and it is also difficult to capture the implicit mechanical degradation that only appears in high load working conditions, resulting in high maintenance cost and difficult to eliminate safety hazards. Therefore, how to realize the spatio-temporal alignment and deep fusion of heterogeneous data, and then accurately decouple the control load and improve the accuracy and robustness of state evaluation, has become a technical problem to be solved. SUMMARY

[0003] To solve the above technical problems, the present application provides a helicopter fleet operation safety state evaluation system based on multi-source data fusion. Specifically, the technical scheme of the present application includes:

[0004] The processor and memory of the on-board sensor and maintenance terminal communication, the processor executes instructions to realize:

[0005] Data alignment: obtaining flight parameters and maintenance logs, cutting flight parameters into action primitives, and analyzing maintenance logs to generate maintenance attention factor numerical weight for specific physical parameters;

[0006] Residual solution: in response to the action type of the action primitive, calling the theoretical baseline from the preset working condition-stress database, calculating the difference between the measured value of the flight parameter and the theoretical baseline, and weighting the difference using the maintenance attention factor numerical weight to obtain the situational health residual;

[0007] Closed loop decision: comparing the situational health residual with the preset safety threshold; if the residual is greater than the safety threshold, generating a high load task prohibition instruction and writing it into the scheduling library; if the residual is not greater than the safety threshold, generating a low load task permission instruction and updating the scheduling library state.

[0008] Preferably, the step of constructing the working condition-stress database comprises: collecting flight parameters with no failure history; clustering standard flight motion primitives by using K-Means algorithm; calculating the arithmetic mean and standard deviation of physical parameters in each mode, taking the arithmetic mean as the theoretical baseline, and taking the standard deviation for setting the floating range of the safety threshold.

[0009] Preferably, the logic of generating the numerical weight of the maintenance attention factor comprises: identifying the target component and operation type in the maintenance log; searching for the corresponding affected physical parameter in the preset association table; assigning a maintenance attention weight greater than one to the flight parameters belonging to the affected parameter, and assigning a reference value equal to one to the flight parameters other than the affected physical parameter.

[0010] Preferably, the closed-loop judgment further comprises implicit degradation detection: simultaneously monitoring the first residual under the steady flight motion primitive and the second residual under the maneuvering motion primitive; only when the first residual is not greater than the steady threshold and the second residual is greater than the maneuvering threshold, an implicit degradation alarm signal is generated; otherwise, the current state flag is maintained.

[0011] Preferably, it further comprises a break-in period adjustment logic: when a component replacement event is identified, a break-in period time window is defined; within the window, the safety threshold is adjusted to 1.2 to 1.5 times of the original value; when the system clock exceeds the end time of the window, the safety threshold is forcibly reset to the original value.

[0012] Preferably, it further comprises maintenance quality verification logic: recording the residual values of the consecutive flight cycles after the maintenance event and calculating the slope over time; if the slope is negative, a maintenance effective confirmation report is generated; if the slope is non-negative, a maintenance quality review instruction is generated.

[0013] Preferably, the method for solving the frequency mismatch of multi-source heterogeneous data in the data alignment step specifically comprises:

[0014] identifying the first sampling frequency of the flight parameter data and the second update frequency of the maintenance log, wherein the first sampling frequency is higher than the second update frequency;

[0015] using zero-order hold interpolation method, the numerical weight of the maintenance attention factor generated based on the low-frequency maintenance log is broadcasted and copied to each high-frequency timestamp corresponding to the flight parameter data along the time axis, so as to realize point-by-point weighted calculation.

[0016] Preferably, the step of generating the numerical weight of the maintenance attention factor for a specific physical parameter by analyzing the maintenance log specifically comprises:

[0017] extracting the component name keywords in the maintenance log by using a predefined regular expression;

[0018] mapping the component name keywords to the preset ATA chapter number;

[0019] Based on the ATA chapter number index to the pre-configured sensor channel ID list, it is determined which physical parameters require the application of maintenance attention factor numerical weight.

[0020] Compared with the prior art, the present application has the following beneficial effects:

[0021] 1. The system effectively decouples the pilot's operation intention and the physical response of the mechanical system by constructing action primitives and dividing flight parameters into independent semantic segments, combined with situational health residual calculation model; this technical means solves the problem of high false alarm rate caused by the inability to distinguish between normal control load and abnormal deviation caused by failure in traditional methods under severe helicopter maneuvering or load mutation, and realizes accurate separation and monitoring of mechanical health status under dynamic working conditions;

[0022] 2. The system generates maintenance attention factor numerical weight by analyzing unstructured maintenance logs, and solves the mismatching problem of low-frequency maintenance data and high-frequency flight parameters in time frequency by using zero-order hold interpolation method; this mechanism breaks the information island between maintenance text data and sensor numerical data, converts descriptive maintenance records into mathematical weighted parameters in monitoring algorithm, so that the system has prior knowledge attention ability, can specifically amplify the parameter weight of recent maintenance components, and significantly improves the early capture ability of maintenance-induced failure or secondary failure;

[0023] 3. The system uses a double-thread parallel monitoring mechanism to simultaneously calculate the residual in steady flight and the residual in dynamic maneuvering; by using the nonlinear characteristics of helicopter dynamics, the system can identify latent failures that behave normally in low-load steady flight and only excite abnormal vibration in high-load dynamic state; this implicit degradation detection logic breaks through the limitations of traditional single threshold monitoring, avoids sudden failures of helicopters during high-load tasks, and thus ensures flight safety;

[0024] 4. The system realizes adaptive boundary control based on physical laws through break-in period adjustment logic and maintenance quality verification logic; on the one hand, the safety threshold is dynamically adjusted to be higher during the break-in time window after component replacement, avoiding false alarms caused by the initial physical characteristics of new components; on the other hand, the residual change slope of consecutive flight cycles is used as objective evidence to realize automatic audit of maintenance quality; this reverse closed loop from monitoring maintenance needs to verifying maintenance quality effectively prevents ineffective maintenance caused by human error, reduces maintenance cost and improves fleet efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0025] The present application will be further explained in conjunction with the accompanying drawings and embodiments:

[0026] Figure 1This is a structural diagram of the system of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0028] Example 1:

[0029] Please see Figure 1 A helicopter fleet operation safety status assessment system based on multi-source data fusion includes a processor and memory that communicate with airborne sensors and maintenance terminals. The processor executes instructions to achieve:

[0030] Data alignment: acquire flight parameters and maintenance logs, segment flight parameters into action primitives, and parse maintenance logs to generate numerical weights of maintenance concern factors for specific physical parameters;

[0031] Residual calculation: In response to the action type of the action element, the theoretical baseline is called from the preset working condition-stress database, the difference between the measured value of the flight parameter and the theoretical baseline is calculated, and the difference is weighted by the numerical weight of the maintenance concern factor to obtain the contextualized health residual.

[0032] Closed-loop determination: The contextualized health residual is compared with the preset safety threshold; if the residual is greater than the safety threshold, a high-load task prohibition instruction is generated and written to the scheduling library; if the residual is not greater than the safety threshold, a low-load task allowment instruction is generated and the scheduling library status is updated.

[0033] This embodiment constructs the physical and logical architecture of the system; the helicopter fleet operation safety status assessment system based on multi-source data fusion has a physical layer including airborne sensors deployed on helicopters, ground maintenance terminals, and core data processing servers;

[0034] At the logical operation level, the processor is configured to execute the following core steps, forming a complete closed loop from data input to decision output:

[0035] Data alignment steps:

[0036] The core of this step lies in resolving the semantic and frequency mismatch problem of heterogeneous data;

[0037] Action primitives refer to independent semantic segments into which continuous time-series flight data are cut according to flight mechanical characteristics. Their function is to provide a unified working condition background for subsequent physical parameter comparison. Their source is pattern recognition of flight parameters collected in real time.

[0038] The numerical weight of maintenance concern factors refers to the coefficients generated based on unstructured maintenance text to quantify the potential risk level of specific mechanical components. Its function is to amplify the weight of parameters affected by recent maintenance events when calculating residuals. Its source is parsed maintenance log text.

[0039] Residual solution steps:

[0040] To accurately identify the true mechanical health status under dynamic operating conditions, this embodiment introduces a contextualized health residual calculation model; this model is not a simple numerical subtraction, but incorporates weighted logic based on the maintenance context:

[0041] In order to simultaneously capture faults with abnormal increases and decreases in physical parameters, this embodiment uses an absolute residual calculation model: ;

[0042] in, The type of action primitive identified at the current moment is used to index the corresponding theoretical baseline from the database; The specific method for determining this is as follows: collect the current time. Real-time flight status feature vectors: collective pitch, periodic pitch, heading, airspeed, and rate of change of altitude. ,calculate With the database stored Clustering Center of Standard Action Element The Euclidean distance; before calculating the Euclidean distance, in order to eliminate the dominant influence of differences in the dimensions and orders of magnitude of different flight parameters on the distance calculation, the processor performs a real-time flight state feature vector analysis. Perform Z-Score standardization; the standardized vector The calculation is as follows: ;

[0043] in, and These are the global mean vector and global standard deviation vector of all samples in the working condition-stress database, respectively; based on Calculate the Euclidean distance to the standardized values ​​of each cluster center; it should be noted that the cluster centers stored in the preset working condition-stress database are... The system is based on the same global mean. and global standard deviation The values ​​were pre-normalized using Z-Score to ensure the consistency of the dimensions in the distance calculation;

[0044] The action label corresponding to the cluster center with the smallest Euclidean distance is selected as... That is, satisfying: ;

[0045] This minimum distance matching method enables the instantaneous mapping of continuously changing real-time flight parameters to a discrete standard operating condition library. Contextualized health residuals, their physical dimensions and Maintaining consistency; it represents the abnormal offset caused purely by mechanical decay or maintenance issues after eliminating normal operating loads; : Measured flight parameters are the actual values ​​of the helicopter at the current moment. The true physical response; in order to eliminate the interference of high-frequency noise from the sensor on the residual calculation, These are not the original sampled values, but smoothed values ​​after being processed by a sliding window filter; in this embodiment, a window width of [missing value] is used. Instant A moving average filter of seconds is used to preprocess the original signal; Theoretical baseline value; it refers to the standard physical parameter values ​​that a healthy helicopter should exhibit under the currently identified motion primitives; The numerical weight of the maintenance concern factor is a dimensionless scalar; it is used to dynamically amplify the parameter deviations of components that have just undergone maintenance or have fault records.

[0046] Increased vibration levels in helicopters could stem from either aggressive pilot maneuvers or mechanical malfunctions; by introducing... The system achieves decoupling of the control load; by introducing a dimensionless coefficient. By modulating the physical difference, the system makes maintenance risks explicit; it should be noted that the action primitives... The recognition is based on multidimensional feature vectors The global operating condition determination is performed, and the formula above contains... and This involves independent calculations for each specific sensor channel, thereby enabling fine-grained monitoring of different subsystems across the entire machine.

[0047] Closed-loop determination steps:

[0048] The system will calculate Compare with the preset safety threshold:

[0049] like If the current fleet members are determined to be in a sub-healthy state, the processor generates a high-load task prohibition instruction and writes it into the scheduling library;

[0050] like If the status is deemed good, a low-load task allow instruction or a full task release instruction is generated, and the scheduling library status is updated.

[0051] This embodiment successfully decouples the pilot's operational intent from the physical response of the mechanical system by introducing action primitives as an intermediate layer, solving the problem of high false alarm rate during violent maneuvers in traditional methods. At the same time, by introducing maintenance concern factors, the information silos between maintenance data and flight data are broken down, and refined task matching is achieved under faulty operation, thereby maximizing the fleet's deployment efficiency while ensuring safety.

[0052] Example 2:

[0053] The steps for constructing the working condition-stress database are as follows: collect historical flight parameters without faults; use the K-Means algorithm to cluster standard flight action primitives; calculate the arithmetic mean and standard deviation of physical parameters under each mode, use the arithmetic mean as the theoretical baseline, and use the standard deviation to set the floating range of the safety threshold.

[0054] This embodiment details the construction process of the working condition-stress database, which is the cornerstone of the system's ability to identify normal values;

[0055] The construction process follows the following rigorous logic:

[0056] Data acquisition: Select flight parameter data from the historical records that are marked as fault-free and have been manually confirmed to be in good condition throughout the entire flight, and construct a golden dataset;

[0057] Action clustering: The K-Means clustering algorithm is used for unsupervised learning on the Golden dataset; the input feature vector includes collective distance, periodic range, heading, airspeed, and rate of change of altitude. The algorithm clusters continuous data into Discrete standard flight maneuver primitives, such as hovering, level flight, left turn, right turn, climb, descent, etc.

[0058] Statistical modeling: For each clustered action primitive Calculate the statistical characteristics of all key physical parameters under this mode:

[0059] Theoretical baseline ( ): Calculate the arithmetic mean; ;

[0060] in, To cluster to the 1st The total number of samples under each flight maneuver mode; For the first in this mode Physical parameter values ​​of each sample; safety threshold fluctuation range ( The calculation method for () is: calculate the standard deviation;

[0061] In this embodiment, the theoretical baseline is directly taken as... The safety threshold used to determine the residual Set as the allowable deviation limit, i.e., take a value of ;

[0062] By constructing a working condition-stress database through K-Means clustering, this invention abandons the high computational requirements of traditional physical modeling and instead establishes a data-driven digital twin in the data space. This method can capture the nonlinear parameter distribution characteristics under complex aerodynamic disturbances, providing a high-precision dynamic reference system for residual calculation.

[0063] Example 3:

[0064] The logic for generating maintenance concern factor weights is as follows: identify the target component and operation type in the maintenance log; find the corresponding affected physical parameter in the preset association table; assign a maintenance concern weight greater than one to flight parameters that are affected parameters, and assign a baseline value equal to one to flight parameters other than affected physical parameters.

[0065] The specific steps involved in parsing maintenance logs to generate numerical weights for maintenance concern factors targeting specific physical parameters include:

[0066] Extract component name keywords from maintenance logs using predefined regular expressions;

[0067] Map component name keywords to preset ATA chapter numbers;

[0068] The system indexes the ATA chapter number to a pre-configured list of sensor channel IDs to determine which physical parameters require the application of maintenance concern factor numerical weights.

[0069] Specifically, to accurately extract operation types and objects, the predefined regular expressions are constructed as structures containing named capture groups; for example, for operations such as replacement, cleaning, inspection, and tightening, the following patterns are constructed:

[0070] Replace, clean, check, and tighten: ; in, and These are named capture groups used for matching operation type keywords and component name keywords, respectively. This indicates the Unicode encoding range that matches any Chinese characters. For example, for an unstructured maintenance log text "2023-10-15 Completed tail reducer magnetic plug cleaning and seal replacement work", the processor uses the above regular expression logic to match the operation type keyword "replacement" and extract the object keywords "seal ring" and "tail reducer". When establishing the mapping, if the tail reducer is identified as a high-level component, the system maps it to the ATA65 section, thereby locking the vibration sensor ID and lubricating oil temperature sensor ID under the system.

[0071] This embodiment focuses on describing how to bridge the semantic gap between text and signal, that is, the specific implementation path for parsing maintenance logs to generate numerical weights of maintenance concern factors;

[0072] The process involves three cascaded mapping steps:

[0073] Keyword extraction: The system uses predefined regular expressions to scan unstructured maintenance logs; it identifies keywords such as tail rotor and replacement.

[0074] Standard coding mapping: The identified component name keywords are mapped to the ATA chapter number of the aviation industry standard; this step eliminates ambiguity caused by different writing habits of maintenance personnel;

[0075] Parameter association and weighting: Based on the ATA chapter number, index to the system's preset sensor channel ID association table; for example, ATA64 is associated with the tail drive shaft vibration sensor and the tail gearbox temperature sensor;

[0076] Weighting:

[0077] For the affected physical parameters in the association table, the system assigns maintenance concern factor numerical weights. In this embodiment, The preferred value range is 1.2 to 2.0; more specifically, the system can assign values ​​in stages according to the depth of the maintenance operation: when the operation type is 'inspection' or 'tightening', Set to 1.2; when the operation type is 'parts replacement' or 'disassembly and repair', Set to 1.5;

[0078] For other irrelevant parameters, assign numerical weights to the maintenance concern factors. ;

[0079] This solution creatively transforms low-frequency, descriptive maintenance texts into mathematical parameters in high-frequency monitoring algorithms. This enables the system to focus on prior knowledge—that is, the system knows where the aircraft has just been repaired, and therefore can more specifically monitor parameter changes in relevant parts, thereby significantly improving the ability to detect faults introduced by maintenance in their early stages.

[0080] Example 4:

[0081] The specific methods for resolving frequency mismatches in multi-source heterogeneous data during the data alignment process include:

[0082] Identify the first sampling frequency of flight parameter data and the second update frequency of maintenance logs, wherein the first sampling frequency is higher than the second update frequency;

[0083] The zero-order hold interpolation method is used to broadcast and copy the numerical weights of maintenance concern factors generated based on low-frequency maintenance logs along the time axis to each high-frequency timestamp corresponding to the flight parameter data, so as to achieve point-by-point weighted calculation.

[0084] This embodiment solves the problem of mismatch in the time dimension of multi-source heterogeneous data, namely the frequency adaptation method in data alignment;

[0085] The specific implementation steps are as follows:

[0086] Frequency identification: The system identifies that the flight parameter data has a first sampling frequency. The maintenance logs generate data with a second update frequency. ;

[0087] Zero-order preserving interpolation:

[0088] In order to make the formula It can be calculated point by point, but the scalar must be... Expanded into a time series of the same length as the flight data; where, The weighted instantaneous residual sequence, This is a real-time sampling sequence of flight parameters. For the corresponding theoretical baseline sequence, This is the weighted sequence after time-axis expansion;

[0089] The system performs the following operations: ;

[0090] in, It is the first The weight generated by the secondary maintenance event This is the repair completion time. This time interval represents the current moment. In the first The end of the next maintenance To the The end of the next maintenance Between; for For specific time periods, or for components for which no relevant records were found in the maintenance log, the system defaults to setting the weight of the maintenance concern factor value. This indicates a baseline state with no maintenance impact; this means that before the next maintenance event occurs, the current weight value is broadcast and copied to the timestamp of every high-frequency flight data point; if multiple maintenance concern events that have not expired exist within the same time period, the system uses the maximum value principle to determine the final weight, i.e. To ensure coverage of the highest-risk items;

[0091] By employing zero-order hold interpolation, this invention achieves time-domain alignment between extremely low-frequency and extremely high-frequency data with minimal computational cost. This not only preserves the continuous influence of the maintenance status but also avoids the destruction of the original spectral characteristics of high-frequency flight parameters by complex resampling algorithms, ensuring the purity of subsequent signal analysis.

[0092] Example 5:

[0093] The closed-loop determination also includes latent decay detection: simultaneously monitoring the first residual under the level flight action element and the second residual under the maneuver action element; generating a latent decay alarm signal only when the first residual is not greater than the steady-state threshold and the second residual is greater than the maneuver threshold; otherwise, maintaining the current state label.

[0094] This embodiment introduces latent decay detection logic to discover latent faults that are normal under normal conditions but become abnormal when subjected to force.

[0095] The system runs two parallel monitoring threads simultaneously:

[0096] Thread A: Activated only when a level flight motion primitive is detected, calculates the first residual. ;

[0097] Thread B: Activated only when a maneuvering primitive is detected, to calculate the second residual. ;

[0098] The decision logic is as follows:

[0099] Among them, steady-state threshold With the mobility threshold Based on the currently identified motion element type, the system retrieves the corresponding motion mode's safety threshold from the working condition-stress database. ; ;

[0100] in, Steady-state threshold, which is usually set quite strictly; The maneuver threshold is usually set quite wide.

[0101] This technology overcomes the limitations of traditional single-threshold monitoring; it utilizes the nonlinear characteristics in helicopter dynamics—some mechanical damage is not apparent under low loads, but only excites abnormal vibrations under high loads; this embodiment can accurately capture this latent decay, avoiding sudden failures of helicopters when performing high-risk missions.

[0102] Example 6:

[0103] It also includes break-in period adjustment logic: when a component replacement event is detected, a break-in period time window is defined; within the window, the safety threshold is increased to 1.2 to 1.5 times the original value; when the system clock exceeds the end time of the window, the safety threshold is forcibly reset to the original value.

[0104] This embodiment describes the break-in period adjustment logic, which is a knowledge-driven adaptive boundary control.

[0105] When the data alignment step detects a component replacement event, the processor executes the following process:

[0106] Define time window: Set the time window from the end of maintenance. The beginning The duration is the break-in period window;

[0107] Threshold Dynamic Increase: Within this window, the system automatically adjusts the safety threshold for the relevant parameters of this component to: ;

[0108] in, It is the break-in factor, with a value of [value missing]. to Then a forced reset is performed: when the system clock... At that time, forced Reset to It should be noted that during the break-in period... Internally, to avoid mathematical cancellation with the weighting logic in Example 3, the processor is configured to temporarily suspend the weighting function of maintenance concern factors, and only adjust... To adapt to the initial physical break-in process of mechanical parts; after the break-in period, restore The computational logic has entered a regular monitoring state; here, it is necessary to further define the timing of the logic switch: within the break-in period window. Internally, the system prioritizes the physical break-in adaptation logic, at which point a forced command is executed. relying solely on Perform tolerance monitoring; when the time exceeds After the physical break-in period, the system automatically switches back to the weighted logic described in Example 3; at this time, the system resets the safety threshold to the original standard. If at the current moment Still a maintenance concern factor During the effective period, it will be restored simultaneously. The weighted logic; to prevent false alarms caused by residual jumps due to simultaneous threshold reduction and weight recovery, the system introduces a smooth transition function at the moment of switching or requires continuous transition. An over-limit alarm can only be triggered after each action element is confirmed; that is, when the system clock... When this happens, the system will forcibly reset the safety threshold to the original standard. Simultaneously, the suspension status of the maintenance concern factors is lifted, and the use of the method described in Example 4 is restored. Weighted calculations are performed to ensure both the restoration of normal monitoring and the monitoring of assembly quality in the later stages, without conflict between the two on the timeline.

[0109] Newly replaced mechanical parts typically generate slightly higher temperatures or particulate content during the break-in period compared to the stable period, which is a natural physical phenomenon. Traditional fixed threshold methods can lead to false alarms caused by such normal phenomena. This embodiment significantly reduces the false alarm rate and unnecessary downtime for inspection by dynamically relaxing the break-in period threshold, demonstrating the system's respect for and adaptation to physical laws. It is worth noting that the maintenance concern factor in Embodiment 3 aims to make the potential maintenance-induced failure risk explicit, while the break-in period adjustment in this embodiment aims to adapt to the physical break-in characteristics of mechanical parts in the early stages. Both are applied to different engineering scenarios and time windows, together forming a complete life-cycle monitoring closed loop.

[0110] Example 7:

[0111] It also includes maintenance quality verification logic: record the residual values ​​of consecutive flights after a maintenance event and calculate the slope of the change over time; if the slope is negative, generate a maintenance validity confirmation report; if the slope is non-negative, generate a maintenance quality review instruction.

[0112] This embodiment demonstrates the unexpected synergistic effect generated by the system—the maintenance quality verification logic;

[0113] The system records maintenance events continuously. The residual value sequence of each flight sortie The preferred range of values ​​for N is... ;like If the value is too small, data fluctuations may cause inaccurate slope calculations; if If the value is too large, the immediacy requirement for maintenance quality verification cannot be met; in this embodiment, it is preferable to set... That is, to use the data from the first 5 flights after maintenance to quickly verify the trend convergence;

[0114] To assess the break-in status after repair, the processor calculates the absolute value of each element in the sequence. Construct the absolute residual sequence The least squares method was used to perform a linear fit on the absolute residual sequence, and the slope was calculated during the linear fit. Previously, the processor used the Laida criterion to remove absolute residual sequences. Outliers are identified to prevent trend distortion caused by spike noise from a single sensor reading; the slope is calculated based on the cleaned sequence. ;

[0115] in, For continuous The arithmetic mean of the flight sortie numbers, For this The arithmetic mean of the absolute residuals corresponding to each flight; For the first Flights The corresponding absolute residual value;

[0116] Calculate the sequence with flight count The slope of the changing trend : ;

[0117] in, This represents the change in the residual, specifically the change in the ordinate of the residual value over a given flight period. This represents the change in the number of flights, specifically the change on the x-axis of the number of flights.

[0118] The decision logic is as follows:

[0119] like This indicates that the residual gradually decreases with the increase of flight sorties, and the system state tends to converge; the processor generates a valid maintenance confirmation report; this means that the new component is running in well and entering a stable state.

[0120] like This indicates that the residual has not improved or is even worsening; the processor generates a repair quality review instruction; this may mean improper installation, improper dynamic balancing, or the use of defective parts.

[0121] This embodiment realizes a reverse closed loop from monitoring maintenance needs through flight parameters to verifying maintenance quality through flight parameters; it uses flight data as objective third-party evidence to automatically audit the operational quality of maintenance personnel; this mechanism can capture invalid maintenance or secondary failures caused by human error, achieving synergistic effects of 1+1>2.

[0122] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A helicopter fleet operation safety status assessment system based on multi-source data fusion, characterized in that, Includes a processor and memory for communicating with airborne sensors and maintenance terminals; the processor executes instructions to achieve: Data alignment: acquire flight parameters and maintenance logs, segment flight parameters into action primitives, and parse maintenance logs to generate numerical weights of maintenance concern factors for specific physical parameters; Residual calculation: In response to the action type of the action element, the theoretical baseline is called from the preset working condition-stress database, the difference between the measured value of the flight parameter and the theoretical baseline is calculated, and the difference is weighted by the numerical weight of the maintenance concern factor to obtain the contextualized health residual. Closed-loop determination: Compare the contextualized health residual with the preset safety threshold; if the residual is greater than the safety threshold, generate a high-load task prohibition instruction and write it into the scheduling library; If the residual is not greater than the safety threshold, generate a low-load task permission instruction and update the scheduling library status.

2. The helicopter fleet operation safety status assessment system based on multi-source data fusion according to claim 1, characterized in that, The steps for constructing the working condition-stress database are as follows: collect historical flight parameters without faults; use the K-Means algorithm to cluster standard flight action primitives; calculate the arithmetic mean and standard deviation of physical parameters under each mode, use the arithmetic mean as the theoretical baseline, and use the standard deviation to set the floating range of the safety threshold.

3. The helicopter fleet operation safety status assessment system based on multi-source data fusion according to claim 1, characterized in that, The logic for generating maintenance concern factor weights is as follows: identify the target component and operation type in the maintenance log; find the corresponding affected physical parameter in the preset association table; assign a maintenance concern weight greater than one to flight parameters that are affected parameters, and assign a baseline value equal to one to flight parameters other than affected physical parameters.

4. The helicopter fleet operation safety status assessment system based on multi-source data fusion according to claim 1, characterized in that, The closed-loop determination also includes latent decay detection: simultaneously monitoring the first residual under the level flight action element and the second residual under the maneuver action element; generating a latent decay alarm signal only when the first residual is not greater than the steady-state threshold and the second residual is greater than the maneuver threshold; otherwise, maintaining the current state label.

5. The helicopter fleet operation safety status assessment system based on multi-source data fusion according to claim 1, characterized in that, It also includes break-in period adjustment logic: when a component replacement event is detected, a break-in period time window is defined; within the window, the safety threshold is increased to 1.2 to 1.5 times the original value; when the system clock exceeds the end time of the window, the safety threshold is forcibly reset to the original value.

6. The helicopter fleet operation safety status assessment system based on multi-source data fusion according to claim 1, characterized in that, It also includes maintenance quality verification logic: record the residual values ​​of consecutive flights after a maintenance event and calculate the slope of the change over time; if the slope is negative, generate a maintenance validity confirmation report; if the slope is non-negative, generate a maintenance quality review instruction.

7. The helicopter fleet operation safety status assessment system based on multi-source data fusion according to claim 1, characterized in that, The specific methods for resolving frequency mismatches in multi-source heterogeneous data during the data alignment process include: Identify the first sampling frequency of flight parameter data and the second update frequency of maintenance logs, wherein the first sampling frequency is higher than the second update frequency; The zero-order hold interpolation method is used to broadcast and copy the numerical weights of maintenance concern factors generated based on low-frequency maintenance logs along the time axis to each high-frequency timestamp corresponding to the flight parameter data, so as to achieve point-by-point weighted calculation.

8. The helicopter fleet operation safety status assessment system based on multi-source data fusion according to claim 1, characterized in that, The specific steps involved in parsing maintenance logs to generate numerical weights for maintenance concern factors targeting specific physical parameters include: Extract component name keywords from maintenance logs using predefined regular expressions; Map component name keywords to preset ATA chapter numbers; The system indexes the ATA chapter number to a pre-configured list of sensor channel IDs to determine which physical parameters require the application of maintenance concern factor numerical weights.

Images