System and method for monitoring aircraft status

Abstract

Methods and systems for monitoring aircraft status are provided. One method includes acquiring a primary value indicative of an aircraft status using a first sensor, generating a preliminary composite value indicative of the aircraft status using other data acquired from a second sensor, the second sensor being of a different operating principle than the first sensor. A computer-implemented model is used to calculate a correction value to be applied to the preliminary composite value, the computer-implemented model having been trained with machine learning and training data that associates previous values of the other data with previous correction values. The correction value is applied to the preliminary composite value to generate a corrected composite value indicative of the aircraft status. The corrected composite value can be used as a basis for controlling the aircraft when the primary value is deemed unreliable or unavailable.

Description

Technical Field

[0001] This disclosure relates generally to aircraft, and more particularly to monitoring the status of aircraft and operating aircraft. Background Technology

[0002] In modern fly-by-wire aircraft, the aircraft's air data system generates air data parameters that are critical to safety. These parameters can be used by the aircraft's flight control system or relied upon by the crew operating the aircraft. Traditional methods for providing the required reliability in such safety-critical systems involve employing multiple redundant measurement channels operating in parallel. These redundant channels introduce design and manufacturing variations and independence between channels. Combined with fault detection and isolation logic, this redundancy provides data availability, and dissimilarity provides data integrity. Unfortunately, this approach is disadvantageous in terms of weight, power requirements, size, installation effort, development complexity, and cost. Furthermore, this approach does not necessarily cover all common failure modes, so channels may be susceptible to similar failure causes. Improvements are desired. Summary of the Invention

[0003] In one aspect, this disclosure describes a method for operating an aircraft. The method includes: During the flight of the aircraft, key aerodynamic data are acquired using sensors that interact with airflow outside the aircraft, and the key aerodynamic data indicates the state of the aircraft. Using non-aerodynamic data from sources other than the aforementioned sensors, preliminary synthetic data indicating the aircraft's status is generated; The computer-implemented model is used to generate a correction for the initial synthetic data. The computer-implemented model has been trained using machine learning and training data that correlates previous non-aerodynamic data with previous corrections. The correction is applied to the initial synthetic data to generate corrected synthetic data that indicates the state of the aircraft; It was determined that the key aerodynamic data was unreliable or unavailable; and After determining that the primary aerodynamic data is unreliable or unavailable, actions are performed on the aircraft based on the corrected synthetic data.

[0004] The action may include controlling the flight control surface of the aircraft.

[0005] The method may include: controlling the flight control surfaces of an aircraft based on the primary aerodynamic data before determining that the primary aerodynamic data is unreliable or unavailable.

[0006] The method may include: when it is determined that the primary aerodynamic data is unreliable or unavailable, comparing the primary aerodynamic data with the corrected synthetic data.

[0007] The action may include transmitting the corrected synthetic data to the aircraft's crew.

[0008] Applying the correction to the preliminary synthetic data may include adding the correction value to the synthetic value of the preliminary synthetic data.

[0009] Applying the correction to the preliminary synthesized data may include multiplying the correction value by the synthesized value of the preliminary synthesized data.

[0010] The method may include: before applying the correction to the preliminary synthesized data: It is determined that the correction value exceeds a specified value; and Adjust the correction value to the specified value.

[0011] Generating the correction may include: using the preliminary synthetic data to identify the operating points of the aircraft within the operating envelope of the aircraft; and calculating the correction based on the operating points within the operating envelope.

[0012] The aircraft status may include the aircraft's airspeed.

[0013] The aircraft state may include the Mach number at which the aircraft is traveling.

[0014] The aircraft status may include the aircraft's angle of attack.

[0015] The aircraft state may include the aircraft's sideslip angle.

[0016] The aircraft status may include the dynamic pressure of the air the aircraft is traveling through.

[0017] The aircraft status may include the temperature of the air the aircraft is traveling through.

[0018] The aircraft status may include the aircraft's pressure altitude.

[0019] The aircraft status may include the aircraft's vertical speed.

[0020] The aircraft status may include the air density outside the aircraft.

[0021] The aircraft status may include the aircraft's weight.

[0022] Implementation examples may include combinations of the features described above.

[0023] In another aspect, this disclosure describes a method for monitoring the status of an aircraft. The method includes: Use the first sensor to acquire key values ​​indicating the aircraft's status; A preliminary composite value indicating the aircraft's status is calculated using other data acquired via a second sensor, the operating principle of which differs from that of the first sensor; A computer-implemented model is used to calculate correction values ​​to be applied to the initial synthesized values; the computer-implemented model has been trained using machine learning and training data that correlates previous values ​​of the other data with previous correction values; and The correction value is applied to the initial composite value to generate a corrected composite value indicating the aircraft's state.

[0024] The key value can be obtained when the first sensor interacts with the airflow outside the aircraft.

[0025] The computer-implemented model may include an artificial neural network. The method may include training the artificial neural network using previous values ​​of the other data associated with the previous corrected values ​​before generating the initial synthesized values.

[0026] The method may include: Verify the stated primary value; and After generating the initial synthesized value, the artificial neural network is further trained using the machine learning and a new correction value based on the difference between the primary value and the initial synthesized value.

[0027] The method may include applying the correction value to the initial synthesized value before: It is determined that the correction value exceeds a specified value; and Decrease the correction value.

[0028] Calculating the correction value may include: Identify the operational points of the aircraft within the operational envelope of the aircraft; and The correction value is calculated based on the aircraft's operating point within its operating envelope.

[0029] The aircraft status may include one or more of the following: the aircraft's airspeed; the aircraft's Mach number; the aircraft's angle of attack; the aircraft's sideslip angle; the dynamic pressure of the air the aircraft passes through; the temperature of the air the aircraft passes through; the aircraft's pressure altitude; air density; the aircraft's weight; and the aircraft's vertical speed.

[0030] Other data acquired by the second sensor may include multiple values ​​acquired at different times.

[0031] Implementation examples may include combinations of the features described above.

[0032] In another aspect, this disclosure describes a computer program product for operating an aircraft, the computer program product including a non-transitory computer-readable storage medium containing program code that can be read / executed by a computer, processor or logic circuit to perform the methods disclosed herein.

[0033] In another aspect, this disclosure describes an aircraft system comprising: Sensors, which are used to acquire key data indicating the status of the aircraft; One or more data processors; and A non-transitory machine-readable storage device, wherein the non-transitory machine-readable storage device stores: An estimator function that can operate to generate preliminary synthetic data based on data acquired from sources other than the sensor; A corrector function operable to generate corrections to be applied to the initial synthetic data using a computer-implemented model trained using machine learning and training data that correlates previous corrections with other previous data; and Instructions, which can be executed by the one or more data processors and are configured to cause the one or more data processors to perform the following operations: The estimator function is used to generate the preliminary synthetic data indicating the state of the aircraft using the other data; The corrector function is used to generate the correction to be applied to the initial synthesized data; The correction is applied to the initial synthetic data to generate corrected synthetic data indicating the aircraft's state; and Generate an output indicating the corrected synthetic data.

[0034] The sensor can be configured to be exposed to airflow outside the aircraft during operation of the sensor.

[0035] The sensor may be a first sensor. The system may include sources other than the sensor. The sources other than the sensor may include a second sensor, the second sensor having a different operating principle than the first sensor.

[0036] The instructions can be configured to restrict the application of the correction to the preliminary synthesized data before the one or more data processors apply the correction.

[0037] Limitations on the application of the correction may include: Determine that the magnitude of the correction exceeds a threshold; and Reduce the magnitude of the correction.

[0038] The instructions can be configured to cause the one or more data processors to apply a filter to the correction before applying the correction to the preliminary synthesized data.

[0039] Calculating the correction may include: identifying the aircraft's operating point within the aircraft's operating envelope; and generating the correction based on the aircraft's operating point within the operating envelope.

[0040] The instructions can be configured to cause the one or more data processors to perform the following operations: determine that the primary data is unreliable or unavailable; and, after determining that the primary data is unreliable or unavailable, issue action commands on the aircraft based on the corrected synthetic data.

[0041] Implementation examples may include combinations of the features described above.

[0042] Further details regarding these and other aspects of the subject matter of this application will become apparent from the following detailed description and accompanying drawings. Attached Figure Description

[0043] Now refer to the attached diagram, in which:

[0044] Figure 1 This is a top plan view of an exemplary aircraft including the aircraft condition monitoring system described herein;

[0045] Figure 2 yes Figure 1 A schematic diagram of an exemplary aircraft condition monitoring system for an aircraft, including an exemplary synthetic data system;

[0046] Figure 3 yes Figure 2 A schematic diagram of an exemplary computer for an aircraft condition monitoring system;

[0047] Figure 4 yes Figure 1 A schematic diagram of an exemplary flight control system for an aircraft, including Figure 2 Aircraft condition monitoring system;

[0048] Figure 5A This is a schematic diagram of another exemplary synthetic data system in operational configuration;

[0049] Figure 5B It is in training configuration Figure 5A A schematic diagram of a synthetic data system;

[0050] Figure 6 This is a schematic diagram of an exemplary corrector function implemented as an artificial neural network;

[0051] Figure 7 This is a flowchart of a method for monitoring the status of an aircraft;

[0052] Figure 8 This is a flowchart of the method for operating the aircraft;

[0053] Figure 9 This is a table illustrating an exemplary data structure, demonstrating one method for generating machine learning training data;

[0054] Figure 10 This is a table showing an example machine learning training dataset. Detailed Implementation

[0055] This disclosure describes systems and methods for monitoring the state of one or more aircraft during aircraft operation (e.g., flight), and optionally, the system and methods are also used to control one or more actions of the aircraft (e.g., operating parameters) based on the monitored aircraft state. In some embodiments, the methods and systems can improve the operation of one or more monitoring systems of the aircraft, and / or can improve the operation of one or more control systems of the aircraft. Using the methods and systems can potentially mitigate the sensing redundancy requirements (and reduce equipment costs) of the safety-critical systems of the aircraft by providing a relatively reliable and independent alternative (i.e., backup) backup data source (in the form of synthetic data generated by analysis). When primary data is determined to be unavailable or unreliable, one or more systems of the aircraft can switch to relying on synthetic data to control one or more actions associated with the aircraft. In some embodiments, switching to use synthetic data can allow the aircraft to continue operating in a non-degraded mode (e.g., preventing aircraft 10 from switching to an alternative degraded operating mode with reduced safety margins).

[0056] In some embodiments, artificial intelligence (AI) enhancements can be used to improve the accuracy of synthetic data. For example, the methods and systems described herein can use a model trained with machine learning (ML) to apply corrections to the synthetic data generated for analysis. Instead of generating the synthetic data itself, the ML-trained model can be limited to generating corrections that are within a specified range, or meet or fall below specified limits, and these corrections are applied to the pre-computed synthetic data. In some embodiments, using an ML-trained model as described herein can improve the accuracy of synthetic data and also facilitate aircraft certification work by applicable certification bodies, since the ML-trained model has limited authority in generating synthetic data. Although the following description specifically refers to aircraft, it should be understood that the use of AI enhancements to improve synthetic data described herein can also be used for other applications, including other types of mobile platforms (e.g., autonomous vehicles), industrial processes (e.g., nuclear power plants), and safety-critical medical devices.

[0057] Aspects of various embodiments have been described with reference to the accompanying drawings. The term "connection" can include a direct connection (where two elements are in contact with each other) and an indirect connection (where at least one additional element is located between the two elements). The term "basic" as used herein can be used to modify any quantitative representation that allows for variation without causing a change in the basic function associated with it. Although terms such as "maximize," "minimize," and "optimize" may be used in this disclosure, it should be understood that these terms can be used to refer to improvements, adjustments, and refinements, and are not necessarily strictly limited to maximum, minimum, or optimal.

[0058] Figure 1 This is a top plan view of an exemplary aircraft 10 including the aircraft condition monitoring (optionally control) system 12 described herein. In various embodiments, aircraft 10 can be any type of manned or unmanned aircraft (e.g., drone), such as corporate, private, commercial, and passenger aircraft. For example, aircraft 10 can be a turboprop aircraft, business jet, hybrid wing body aircraft, or jet airliner. Aircraft 10 can be a fixed-wing aircraft as shown herein, or it can be a rotorcraft (e.g., helicopter). Aircraft 10 can include wings 14, fuselage 16, one or more engines 18 for propelling aircraft 10, and actuable control surfaces 20. One or more engines 18 can be mounted to fuselage 16 and / or wings 14. Control surfaces 20 can be actuable aerodynamic devices that allow adjustment and control of the aircraft's altitude. Control surfaces 20 can include one or more primary control surfaces, such as elevators, ailerons, and rudders, for adjusting the pitch, roll, and yaw of aircraft 10, respectively. The control surface 20 may include one or more auxiliary control surfaces, such as spoilers, trailing edge flaps, leading edge slats, horizontal stabilizers, and speed brakes.

[0059] During the operation (i.e., flight) of the aircraft 10, the flight control surface 20 can be actuated based on commands manually entered by the crew relying on one or more aircraft states that can be directly sensed and / or calculated (i.e., synthesized). In some embodiments of the aircraft 10, and in some cases, the flight control surface 20 can be automatically actuated based on commands generated by the flight control system using control laws that also rely on one or more aircraft states that can be directly sensed and / or calculated.

[0060] Figure 2 This is a schematic diagram of an exemplary aircraft condition monitoring system 12 (hereinafter referred to as "monitoring system 12") of aircraft 10. Monitoring system 12 may include one or more master sensors 22 (hereinafter referred to as the singular) operable to acquire key data 24 indicating the aircraft's condition. Key data 24 may be obtained from direct measurements by the master sensors 22. Key data 24 may be provided to one or more consumable systems 26 (hereinafter referred to as the singular) of aircraft 10. Consumption system 26 may include a system that controls one or more devices of aircraft 10 (e.g., flight control surface 20, engine 18) based on key data 24. Consumption system 26 may include electronic displays or other alert devices located in the cockpit of aircraft 10 to transmit key data 24 to the crew when needed. In some embodiments, consumable system 26 may include aircraft 10's flight management system, aircraft 10's autopilot system, aircraft 10's automatic landing system, aircraft 10's automatic throttle system, and / or engine 18.

[0061] The main sensor 22 may be operatively connected to the consumption system 26 so that the main data 24 can be transmitted to the consumption system 26 in a manner that bypasses the computer 28. Alternatively, the main sensor 22 may be operatively connected to the computer 28 so that the main data 24 can be optionally transmitted to the computer 28. In some embodiments, the main data 24 may be transmitted to the consumption system 26 via the computer 28.

[0062] The monitoring system 12 may include one or more other data sources 30 (hereinafter referred to in the singular) (e.g., sensors) operable to acquire additional data 32. The other data sources 30 may be operatively connected to the computer 28 so that the additional data 32 can be transmitted to the computer 28. The additional data 32 may be used as a basis for generating preliminary synthetic data 34 and corrected synthetic data 36 using the synthetic data system 35. The preliminary synthetic data 34 and corrected synthetic data 36 may also indicate the same aircraft state as represented by the primary data 24. The preliminary synthetic data 34 may be computed using one or more estimator functions 38 (hereinafter referred to in the singular) of the synthetic data system 35, which may be implemented by the computer 28. As explained below, one or more corrector functions 40 (hereinafter referred to in the singular) of the synthetic data system 35 may be operable to generate one or more corrections 42 (hereinafter referred to in the singular) to be applied to the preliminary synthetic data 34 using a computer-implemented model that has been iteratively trained using appropriate ML training data (e.g., supervised ML) that correlates previous additional data with previous corrections. In some embodiments, the corrector function 40 may include one or more multidimensional lookup tables or best-fit (e.g., least-squares fit) functions. In some embodiments, the corrector function 40 may be obtained through non-iterative machine learning.

[0063] The corrected synthetic data 36 can provide an alternative (e.g., backup) data source for use by one or more systems of the aircraft 10, and this data source is generated without directly measuring the relevant specific aircraft state. Therefore, the initial synthetic data 34 can be calculated (e.g., derived) from other data 32 that originate from sources independent of and different from the main sensor 22. For example, other data sources 30 may include sensors of a different type and operating principle than the main sensor 22. Therefore, the reliability of the corrected synthetic data 36 may not be affected by the same causes of failure or error as the main sensor 22. In other words, the corrected synthetic data 36 can be generated in a manner substantially completely independent of the main data 24 and can be generated based on different measurement principles and equipment. The corrected synthetic data 36 can potentially reduce operational risks on the aircraft 10 by creating (e.g., analyzing) an additional layer of redundancy for the main data 24. In some cases, using the corrected synthetic data 36 can reduce the redundancy requirements for acquiring the main data 24.

[0064] During normal operation of aircraft 10, primary data 24 may be considered to have a higher level or importance than corrected synthetic data 36 and may be relied upon by consumption system 26 to control one or more actions related to aircraft 10. In other words, during normal operation, primary data 24 may be given higher priority than corrected synthetic data 36. However, when primary data 24 is determined to be unavailable or unreliable, consumption system 26 may switch to relying on corrected synthetic data 36 as an alternative (e.g., backup) data source instead of primary data 24 to control one or more actions related to aircraft 10. In other words, corrected synthetic data 36 can serve as alternative data to allow for mitigation of failures on aircraft 10 that jeopardize primary data 24.

[0065] Different estimator functions 38 can be associated with different aircraft states. In some embodiments, the estimator function 38 can be model-based. The estimator function 38 can be based on a relatively simple analytical model that captures prior knowledge of the behavior and performance of the aircraft 10. The model of the estimator function 38 can take different forms, including algebraic expressions and integral-differential relations. The model of the estimator function 38 can be linear or nonlinear. Depending on the specific aircraft state being monitored, the knowledge embedded in the estimator function 38 can include known laws of flight physics, the kinematics and dynamics of the aircraft 10, and aerodynamic characteristics corresponding to different configurations of the aircraft 10, such as different positions of the flight control surfaces 20 and different positions of the landing gear of the aircraft 10. The knowledge embedded in the estimator function 38 can also include computational fluid dynamics data, wind tunnel test data, simulation and flight experiment data, past experience, equipment accuracy and performance data, etc. In some embodiments, a simpler estimator function 38 can be based on the balance of forces in straight and horizontal flight and stable maneuvers, and implemented in unstable and transient flight according to Newton's second law (i.e., force = mass × acceleration). The model of estimator function 38 can be linear or nonlinear, and may or may not include parameter scheduling as a function of flight conditions, operating points, and / or the configuration of aircraft 10. Estimator function 38 may include dynamic filters, such as complementary filters, Kalman filters, extended Kalman filters, unscented Kalman filters, control loop type estimators, state observers, etc.

[0066] Simple and more complex analytical models capture knowledge of a system, but may not be able to do so completely. Analytical models are approximations of reality. Furthermore, they may not fully capture uncertainties associated with the measuring equipment (e.g., accuracy, delay, drift, offset, scale factor error, etc.) or all disturbances that may arise from air turbulence or other factors. For example, an analytical model may fail to capture nonlinearities such as airflow separation or interference caused by fuselage or unsteady conditions, boundary layer effects, shock waves, and interference with other equipment and systems (e.g., antennas, landing gear and doors, deployed ramjet turbines, extended lift-enhancing devices such as flaps and leading-edge slats). In some cases, properly analyzing and modeling these effects and / or generating an analytical model that fully captures the knowledge of the system being modeled may be impractical or difficult.

[0067] A simpler analysis model that is easier to implement might be preferred. For both simple and complex analysis models, the accuracy of the initial synthetic data 34 may be insufficient to effectively detect and identify faults in key air data parameters, or to provide backup air data that allows the aircraft 10 to maintain its nominal (or alternative) flight control operating mode if the primary data 24 is unreliable or unavailable. Causes of this inaccuracy may include insufficient precision of the analysis model itself, the limited precision of the input measurements themselves (reflecting imperfections in the sensors used to measure them), or the calculations used with these measurements.

[0068] The corrector function 40 can provide AI-based enhancements to compensate for potential inaccuracies in the initial synthetic data 34 generated by the estimator function 38. In some embodiments, the corrector function 40 may include or be based on an ML-trained model trained with ML and training data that correlates previous other data with previous corrections. Therefore, the corrector function 40 can compute a correction 42 based on the other data 32. The correction 42 can then be applied to the initial synthetic data 34 to generate corrected synthetic data 36. In some embodiments, applying the correction 42 to the initial synthetic data 34 may include adding a numerical correction value to the numerical synthetic value of the initial synthetic data 34 at a summation point 44. Alternatively or supplementarily, applying the correction 42 to the initial synthetic data 34 may include multiplying the numerical correction value by the numerical synthetic value of the initial synthetic data 34.

[0069] Instead of directly using an ML-trained model to compute synthetic data, the ML-trained model of the corrector function 40 is used to compute correction 42, which is applied to the preliminary synthetic data 34 computed using the estimator function 38. In other words, the synthetic data system 35 can combine the advantages of ML with the advantages of analytical system knowledge embedded in the estimator function 38. In other words, the synthetic data system 35 provides a hybrid solution that combines analytical data estimation methods with ML-based augmentation methods, leveraging the advantages of both while avoiding their disadvantages. Therefore, the ML-trained model of the monitoring system 12 can have limited authority by providing augmentation rather than providing synthetic data generated entirely by the ML-trained model. As further explained below, ML can be used to learn and compensate for unknown and / or complex analytical knowledge about the aircraft 10 and inaccuracies in the estimator function 38, thereby improving the accuracy of the preliminary synthetic data 34.

[0070] In some cases, using the corrector function 40 allows for the use of a simpler (and less accurate) estimator function 38, while still achieving an acceptable level of accuracy for the corrected synthetic data 36 through correction 42. The simpler estimator function 38 can reduce development time and, in some cases, computational burden. In some embodiments, the corrector function 40 can also compensate for inaccuracies in measurements and / or calculations acquired from sensors via other data sources 30.

[0071] In some embodiments, the monitoring system 12 may include a verification function that determines whether the primary data 24 is reliable or unreliable. In some embodiments, the verification function may implement a voting and arbitration scheme to select the most suitable values ​​for various parameters based on an assessment of their validity and a determination of their accuracy. In some embodiments, the verification function may implement random fault monitoring and / or common-mode fault monitoring to determine the status of different data sources and their parameters.

[0072] In addition to providing a backup for the primary data 24, the preliminary synthetic data 34 and / or the corrected synthetic data 36 can enhance fault detection (e.g., random error and common cause error monitoring) and isolation capabilities in failure events that jeopardize the primary data 24. In some embodiments, the preliminary synthetic data 34 and / or the corrected synthetic data 36 can be used as additional redundant sources in voting and arbitration functions for the divergent sources of the primary data 24, and / or for evaluating the reliability of the primary sensor 22 without the need for additional equipment (e.g., redundant sensors). For example, a verification function may include comparing the primary data 24 with the preliminary synthetic data 34 and / or the corrected synthetic data 36 to determine whether the primary data 24 is reliable or unreliable. In some embodiments, the preliminary synthetic data 34 and / or the corrected synthetic data 36, ​​due to their complete or partial independence from the primary data 24, can increase the likelihood of detecting and isolating common-mode faults. Where the primary data 24 is already provided by two independent sources and different measurement principles, the preliminary synthetic data 34 and / or the corrected synthetic data 36 can be used to arbitrate and select the correct source, for example, by isolating insufficient data and sources.

[0073] In some embodiments, the corrector function 40 may consider the operating point of the aircraft 10 within the operating (e.g., flight) envelope of the aircraft 10. Operating points may include normal aircraft configurations, such as clean cruise, descent, approach, or landing configurations, in which, for example, landing gear is deployed and / or lift enhancement devices are extended. Operating points may include anomalous aircraft configurations resulting from one or more failures, such as engine failure (e.g., one engine not working). The corrector function 40 may be configured (e.g., ML-trained) to perform corrective scheduling based on operating points. For example, the corrector function 40 may use different training models and / or corrective scheduling based on operating points (e.g., based on speed or altitude) and / or based on the current flight phase of the aircraft 10. Other data 32 and / or preliminary synthetic data 34 may indicate operating points and may be used by the corrector function 40 to calculate corrections 42 applicable to the vicinity of that particular operating point. The operating envelope may define operational limits for the aircraft 10 in terms of maximum speed and load factor at a given atmospheric density. The operating envelope can define one or more operating condition ranges within which the aircraft 10 can operate safely. The operating point can be determined based on other data 32 or on preliminary synthetic data 34. In some embodiments, the correction function 40 can identify the operating point of the aircraft 10 within its operating envelope and generate a correction 42 based on that operating point. In some embodiments, a portion of the other data 32 and / or the preliminary synthetic data 34 can be preprocessed, such as subjected to limiting and filtering, before being used as input to the corrector function 40. The correction function 40, using the preliminary synthetic data 34, can help capture the behavioral and performance characteristics of the aircraft 10 under different flight conditions.

[0074] As further described below, the monitoring system 12 can be used to monitor air (i.e., aerodynamic) data and can be integrated into the flight control system 46 of the aircraft 10 (e.g., Figure 4 (As shown). However, monitoring system 12 can also be integrated into other systems of aircraft 10 (e.g., propulsion systems, such as engine 18), where synthetic data can be beneficial. For example, monitoring system 12 can be used in systems with relatively stringent requirements for the availability of data (indicating aircraft status). Monitoring system 12 can be used in safety-critical systems where sufficient primary data 24 and other data 32 are available to allow the generation of corrected synthetic data 36.

[0075] Figure 3 This is a schematic diagram of an exemplary computer 28 of the monitoring system 12. In some embodiments, the computer 28 may be implemented as one or more air data computers 28A (such as...). Figure 4As shown, and hereinafter referred to in the singular as "ADC28") and / or one or more (e.g., master) flight control computers 28B (such as Figure 4 As shown, and hereinafter referred to as "FCC 28" in the singular. Computer 28 (hereinafter referred to as "processor 48") may include one or more data processors 48 (hereinafter referred to as "processor 48") and non-transitory machine-readable storage 50 (hereinafter referred to as "singular"). Computer 28 may be configured to generate one or more outputs, such as corrected synthetic data 36, ​​based on one or more inputs (e.g., other data 32). In some embodiments, computer 28 may optionally perform other tasks not disclosed herein. Computer 28 may perform one or more procedures or steps defined by instructions 52 stored in memory 50 and executable by processor 48 to generate corrected synthetic data 36.

[0076] Processor 48 may include any suitable devices configured to cause computer 28 to perform a series of steps to implement a computer-implemented process, such that instructions 52, when executed by computer 28 or other programmable means, can perform the functions / behaviors specified in the methods described herein. Processor 48 may include, for example, any type of general-purpose microprocessor or microcontroller, digital signal processing (DSP) processor, integrated circuit, field-programmable gate array (FPGA), reconfigurable processor, graphics processing unit (GPU), dedicated neural engine processing unit (NPU), other suitable programmable or programmable logic circuits, or any combination thereof.

[0077] Memory 50 may include any suitable machine-readable storage medium. Memory 50 may include non-transitory controller-readable storage media, such as, but not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any suitable combination thereof. Memory 50 may include any suitable storage means (e.g., devices) adapted to store machine-readable instructions 52 executable by processor 48 in a retrievable manner. In some embodiments, estimator function 38 and corrector function 40 may also be stored in memory 50.

[0078] Figure 4 This is a schematic diagram of an exemplary flight control system 46 (hereinafter referred to as "FCS 46") of the aircraft 10, which integrates... Figure 2 The monitoring system 12. FCS 46 can use (e.g., safety-critical) air data to control one or more actions related to the aircraft 10. The air data can represent the air quality conditions surrounding the aircraft 10 during flight. This air data can be used by the crew and onboard systems to make operational decisions and control actions related to the aircraft 10.

[0079] The FCS 46 may include additional components beyond those described herein. The physical signals carried by the indicated signal lines may be analog or digital. The signal lines may be multidimensional and may carry multiple physical signals, parameters, or data elements. The FCS 46 may include digital signal transmitting and receiving hardware components to implement processing functions such as buffering, encoding, encryption, and verification. The FCS 46 may include one or more analog-to-digital converters.

[0080] The FCS 46 can actuate one or more flight control surfaces 20 using air data as a basis via one or more flight control laws 54 to adjust and control the flight altitude of the aircraft 10. For example, the FCC 28B and the flight control surface 20 can be operatively connected in a fly-by-wire configuration via one or more actuators and controllers so that the FCC 28B can use the air data to generate commands for actuating the flight control surface 20. In some embodiments, the FCC 28B can transmit certain air data to the crew of the aircraft 10 via one or more alarms 56 (e.g., displays). Alternatively or supplementarily, certain air data can be transmitted directly to the consumption system 26 instead of via the FCC 28B. For example, certain air data can be transmitted to the consumption system 26 via a communication path that does not include the FCC 28B. Examples of measured or calculated air data may include airspeed (e.g., calibrated airspeed, vacuum speed), Mach number, angle of attack, sideslip angle, dynamic pressure (e.g., static pressure and total pressure), external air temperature, pressure altitude, vertical velocity (i.e., rate of climb or descent), air density outside the aircraft 10, etc.

[0081] The FCC 28B can communicate with an aerodynamic air data system including a main sensor 22 and an ADC 28A. In this embodiment, the main sensor 22 may be an aerodynamic sensor configured to be exposed to airflow outside the aircraft 10 during operation of the main sensor 22. Basic air data can be measured via a static pressure port and a total pressure probe, which may be combined with a multi-function probe, angular (i.e., pivotable) blades, and a temperature sensor. In various embodiments, the main sensor 22 may include one or more of the following: an air data probe SmartProbe®, a multi-function probe, a pitot tube static pressure probe, a static pressure port, an angle-of-attack blade, a sideslip angle blade, and a total air temperature probe. In some embodiments, the main sensor 22 may include multiple redundant and optionally dissimilar sensors.

[0082] In various embodiments, some or all of the functions of the ADC 28A may be hosted in the FCC 28B, or may be implemented in hardware that is separate from but communicates with the FCC 28B. For example, the ADC 28A may include one or more (e.g., redundant) line-swappable units, and / or one or more integrated modular avionics modules.

[0083] The main sensor 22 may be located in the airflow outside the forward section of the fuselage 16 and may require detailed calibration and compensation to reflect values ​​at a reference point such as the aircraft's center of mass or center of gravity. The main sensor 22 used to measure air data may be adversely affected by environmental conditions, other conditions, or events. For example, each individual main sensor 22 (e.g., a probe) and its associated electronics may potentially be susceptible to random failures independently of other main sensors 22. Furthermore, multiple identical main sensors 22 may potentially be simultaneously affected by design, manufacturing, or installation defects (i.e., hardware and software). Additionally, exposure of the main sensor 22 to the airflow may cause multiple main sensors 22 to suffer similar failures simultaneously, such as from volcanic ash or from unexpected severe icing conditions clogging or contaminating the pressure port. This simultaneous failure due to a common cause is referred to herein as a common-mode failure.

[0084] The air data probe and the measurements acquired through the aerodynamic air data system are referred to as "aerodynamic" because they involve measuring local airflow characteristics such as pressure (total and static pressure), temperature (total and / or static), or direction (angle of attack, sideslip angle). Other air data parameters that can be calculated from these measurements by the ADC 28A include, for example, air velocity (calibrated air velocity, vacuum velocity) and Mach number.

[0085] In some embodiments, the aerodynamic air data system may optionally be supplemented by additional air data signals derived from engine 18. For example, FCC 28B may optionally communicate with an aerodynamic engine data system associated with engine 18 of aircraft 10, which may include one or more air data probes 58. FCS 46 may include direct connections to air data probes 58 or other engine sensors. FCS 46 may include indirect digital connections to air data probes 58 via Full Authority Digital Engine Control (FADEC) 60 of engine 18. These additional air data signals may include full air temperature and measured pressure, or calculated or estimated air data parameters. These additional signals may be used for monitoring or as supplementary (backup) data for FCS 46. These additional signals may be used as part of the input signal management functions described further below.

[0086] In some embodiments, the pneumatic air data system may optionally be supplemented by additional air data signals derived from the optical air data system (including laser unit 62 and laser controller 64) to enhance the robustness of the FCS 46. For example, the FCC 28B may optionally communicate with the optical air data system. The optical air data system can provide a complete set or partial set of air data parameters and may require certain parameters derived from other systems. For example, to extend the operating envelope of the optical air data system above a certain altitude, estimations of altitude or air density may be required. In some embodiments, an advantage of adding the optical air data system to the pneumatic air data system is that the optical air data system may not depend on aerodynamic measurements and therefore will not be affected by simultaneous performance degradation of the pneumatic air data system. Common-mode failures affecting both the pneumatic and optical air data systems can also be avoided. For example, non-standard icing conditions or other probe blockages may not necessarily affect both systems simultaneously. In the event of performance degradation in one system, a third-party arbitration system can be used to identify the reliable system, thereby isolating the faulty (e.g., unreliable) system.

[0087] When integrated into FCS 46, the synthetic data system 35 can serve as an alternative air data system, capable of generating synthetic air data volume without the need for direct air data measurement. In some embodiments, the synthetic data system 35 can potentially further mitigate risk by creating additional (e.g., analytical) redundancy layers for aerodynamic air data systems such as pitot tube hydrostatic probes and pivotable blades. The addition of the synthetic data system 35 can provide additional differential redundancy levels, enhancing safety without compromising equipment weight, cost, and complexity. In some embodiments, the synthetic data system 35 can provide a method for reducing equipment at a similar level of safety compared to conventional systems. The synthetic data system 35 can also be used to detect failures in other subsystems via data compatibility checks. In some cases, the synthetic data system 35 can mitigate the requirements for redundant parallel measurement chains in aerodynamic air data systems under given availability requirements, or avoid the need to add redundant parallel measurement chains to meet more stringent availability requirements, by synthesizing redundant data from readily available data from other (e.g., non-aerodynamic) data sources 30.

[0088] Compared to the aerodynamic data mentioned above, non-aerodynamic data refers to any other data obtainable from other aircraft systems, as well as any other design or performance knowledge that can be used to synthesize air data parameters. Generating air data parameters from both aerodynamic and non-aerodynamic data allows the two sets of air data parameters to be independent and distinct. This is possible because there is consistency (e.g., analytical redundancy) between different data sources based on other knowledge (such as the physics-based flight mechanics / dynamics, aerodynamics, and design of aircraft 10). This consistency is lost if one or more sensors of a given type fail due to a common cause (such as uncontrolled icing conditions or blockage caused by volcanic ash), and the failure can be detected. With independent and distinct devices, the failed (e.g., unreliable) device can be identified and isolated.

[0089] In some embodiments, other data 32 acquired through other (e.g., non-aerodynamic) data sources 30 may be acquired independently of and in a manner different from the air data acquired through the main sensor 22 and the air data acquired through the laser unit 62 of the optical air data system. Other data 32 acquired from other data sources 30 may include inertial reference data, such as aircraft attitude and heading, body speed and acceleration, ground speed and vertical speed, horizontal position and altitude. These other data sources may include the inertial reference system of the aircraft 10, the attitude and heading reference system of the aircraft 10, the inertial measurement unit of the aircraft 10, the global navigation satellite system (GNSS) receiver of the aircraft 10, such as the global positioning system (GPS) receiver, Galileo receiver, GLONASS receiver, BeiDou receiver, and / or other aircraft navigation devices and systems. Other data 32 may also include the aircraft's mass or weight, center of mass or center of gravity position, moment of inertia estimated by the flight management system of the aircraft 10, various engine data such as thrust and fuel consumption rate provided by the engine power system, deflection and rate of flight control surfaces, flight control surface actuator loads measured by FCS46, and aircraft configuration and status information for each system, such as landing gear position (retracted / extended) and the position of lift-enhancing surfaces such as trailing edge flaps and leading edge slats. Other data sources 30 may include one or more sensors that operate on different principles than the main sensor 22 and measure parameters different from those of the main sensor 22. For example, the sensors of other data sources 30 may be non-aerodynamic and therefore may not be directly exposed to the airflow outside the aircraft 10.

[0090] Other data 32 can be used as (e.g., analytical) estimator functions 38 (in... Figure 2The input signal is shown in the figure. The estimator function 38 can generate (i.e., synthesize) one or more estimates of the air data parameters without relying on their direct measurement. The output parameters of the estimator function 38 are called preliminary synthetic data 34 because they are generated by computation and analysis using data other than the primary data 24 acquired by the main sensor 22 of the aerodynamic air data system. The estimator function 38 can combine other data 32 with some estimation algorithms and other system knowledge (such as aircraft kinematics / dynamics and / or aerodynamic models) to generate the desired preliminary synthetic data 34. In some embodiments, the synthetic data system 35 can be implemented as a software function performed by the FCC 28B or ADC 28A. In some embodiments, the functionality of the synthetic data system 35 can be implemented in hardware that is separate from and communicates with the FCC 28B. For example, the synthetic data system 35 can be integrated as part of an integrated modular avionics system or integrated into a host line replaceable unit that communicates with the FCC 28B for data.

[0091] Air data from various systems, such as air data from a pneumatic engine data system, aerodynamic air data system, optical air data system, and / or synthetic data system 35, can be provided to an input signal manager 65 (hereinafter referred to as "ISM 65"). ISM 65 can periodically perform functions in a substantially real-time manner (e.g., at a rate of 100 Hz) as per FCC 28B. ISM 65 may include one or more sub-functions, each performing periodically at a rate equal to or less than that of ISM 65. For example, ISM 65 may include a data processing function 66 operable to receive data, decode and scale data, verify data, filter data to reduce noise and / or remove structural vibrations at specific frequencies, and / or perform coordinate system transformations, including transforming data to a predefined reference point, etc. In some embodiments, ISM 65 may perform a voting and arbitration function 68 operable to select the most suitable value for various received parameters based on an evaluation of the validity of the received parameters and a determination of their accuracy. In some embodiments, the ISM 65 may perform a random fault monitoring function 70 and a common-mode fault monitoring function 72, which are operable to determine the status of different data sources and their parameters.

[0092] Following ISM 65 and optionally one or more other functions and monitoring 74, an optimal dataset can be selected and provided to flight control law 54 and associated flight mode logic. Some or all of the optimal dataset can also be provided to other consumable systems 26, such as flight management system, autopilot, automatic landing and / or automatic throttle system, and / or engine 18 systems, via the data communication network on aircraft 10.

[0093] Figure 5A This is a schematic diagram of another exemplary synthetic data system 135 in an operational (e.g., non-training) configuration. Synthetic data system 135 can be integrated into... Figure 2 The monitoring system 12 is integrated into Figure 4 The FCS 46 and / or integrated into another system of the aircraft 10. The synthetic data system 135 is shown together with the ISM 65 of the FCS 46. The synthetic data system 135 may include elements of the synthetic data system 35 described above. Identical elements are identified by reference numerals increased by 100. The synthetic data system 135 may be... Figure 2 The computer 28 shown is implemented. Other data 32 can be used as the basis for generating preliminary synthetic data 134 and corrected synthetic data 136 using the synthetic data system 135. The preliminary synthetic data 134 and corrected synthetic data 136 can also indicate the same aircraft state represented by the primary data 24. The preliminary synthetic data 134 can be computed using one or more estimator functions 138 (hereinafter referred to in the singular) of the synthetic data system 135. One or more corrector functions 140 (hereinafter referred to in the singular) of the synthetic data system 135 can operate to generate one or more preliminary corrections 142A (hereinafter referred to in the singular) to be applied to the preliminary synthetic data 134. The corrector function 140 can include a computer-implemented model that is iteratively trained using appropriate ML training data (e.g., supervised ML) that correlates previous other data with previous corrections.

[0094] Limiting or defining corrections within specified ranges can constrain the authority of the corrector function 140 to modify the preliminary synthetic data 134 obtained from the estimator function 138. During normal operation of the synthetic data system 135, the magnitude of the correction can be relatively small relative to the applicable value of the preliminary synthetic data 134. For example, in some cases, the magnitude of the correction to be added to or subtracted from the preliminary synthetic data 134 can be less than 10% (e.g., between 0% and 10%) or less than 5% (e.g., between 0% and 5%) of the applicable value of the preliminary synthetic data 134. An excessively large preliminary correction 142A can indicate a problem with the performance of the corrector function 140. Limiting the effective correction 142B applied to the preliminary synthetic data 134 can be used to limit the corruption of the corrected synthetic data 136 to a safe value.

[0095] Constraining the permissions of the corrector function 140 can aid in the certification of the aircraft 10, as defects in the derived corrections will limit their impact and criticality on the operation of the aircraft 10. Therefore, the initial correction 142A can be limited (e.g., delimited) by the limiter 176 and / or filtered by the filter 178 to generate an effective correction 142B that is subsequently applied to the initial synthesized data 134 via addition at the summation node 144. Applying constraints to the initial correction 142A may include determining that the magnitude of the initial correction 142A exceeds a threshold, and reducing the magnitude of the initial correction 142A. Applying constraints to the initial correction 142A may also include determining that the value of the initial correction 142A exceeds a predetermined value, and adjusting the value of the initial correction 142A to the predetermined value. In other words, the limiter 176 may, in certain circumstances, cause the magnitude of the effective correction 142B to be lower than that of the initial correction 142A. Alternatively or supplementarily, the filter 178 may be used to evaluate the initial correction 142A according to one or more criteria and prevent erroneous corrections from being applied to the initial synthesized data 134. Filter 178 can be configured to reduce and / or smooth noise (e.g., rapid and relatively large changes) in the values ​​of the preliminary correction 142A by using a low-pass filter. Filter 178 can be configured to apply larger corrections more progressively according to expected or known aircraft dynamics and related disturbances caused by turbulence and gusts. Further processing of the preliminary correction 142A may also be performed before it is applied to the preliminary synthetic data 134. Applying filtering and / or other processing to the preliminary correction 142A can be beneficial for performance reasons. In some embodiments, the synthetic data system 135 may include an enable / disable function that can disable the corrector function 140 based on logical checks of the availability and / or integrity of other data 32, the magnitude of the preliminary correction 142A or effective correction 142B, and / or the accuracy of various types of data used within the synthetic data system 135.

[0096] The corrected composite data 136 can be provided to ISM 65 and processed as described above. Other values ​​from primary data 24, preliminary correction 142A, and valid correction 142B can optionally be provided to ISM 65 and used to determine the validity status of the corrected composite data 136 in the voting and arbitration function 68. If determined to be invalid, the corrected composite data 136 will not be used, and appropriate notification can be provided to other aircraft systems and / or the crew. In some cases, the (uncorrected) preliminary composite data 134 can be used to support a degraded operating mode for aircraft 10. If determined to be valid, the corrected composite data 136 can, for example, be used as a substitute for primary data 24, and optionally as a substitute for auxiliary / backup data acquired through the optical air data system.

[0097] After processing by ISM 65, appropriate data can be provided to consumption system 26. In some embodiments, ISM 65 can output corrected composite data 136, primary data 24, and data selection 80. Data selection 80 may be the result of voting and arbitration function 68 and / or other processing performed by ISM 65 to indicate to consumption system 26 which air data source should be used to control one or more components of aircraft 10. Using data selection 80, consumption system 26 can select a suitable data source.

[0098] The corrector function 140 may be driven by the same additional data 32 as the estimator function 138. Alternatively, a portion of the additional data 32 used to drive the corrector function 140 may differ from another portion of the additional data 32 used to drive the estimator function 138. The additional data 32 may be independent of the initial synthetic data 34. In some embodiments, the additional data 32 used by the estimator function 138 and the corrector function 140 to synthesize parameters may be independent of the equivalent sensed parameters from the primary data 24. In some embodiments, some primary data 24 may optionally be used by the corrector function 140 and / or the estimator function 138. Alternatively or supplementarily, some of the initial synthetic data 134 generated by the estimator function 138 may be used as input to the correction function 140 to help identify the current operating point in the operating envelope of the aircraft 10 and accordingly calculate the initial correction 142A. However, the output from the corrector function 140 (e.g., the initial correction 142A) may remain independent of the primary data 24 that can be used to calculate the corresponding initial synthetic data 134.

[0099] In some embodiments, air data parameters (e.g., a subset thereof) from the primary data 24 can be used as inputs to the estimator function 138 when they are independent of and not directly related to the parameters to be estimated by the estimator function 138. In this way, different estimates can be generated for monitoring and evaluating the validity of the air data. For example, the impact pressure from the aerodynamic air data system can be used to calculate the composite angle of attack based on a set of other data 32 that are unrelated to the primary data 24. This composite angle of attack can then be compared with an angle of attack obtained by direct measurement using angle-of-attack blades. In another example, the sideslip angle (AOS) can be estimated from the differential static pressure measured on opposite sides of the fuselage of the aircraft 16. The AOS can also be measured directly by vertical blades similar to angle-of-attack blades. The composite value of the AOS can also be estimated from the lateral load factor measured by the inertial reference unit, rudder position, known aircraft aerodynamics, and aircraft weight. Therefore, estimating the composite value of the AOS may also require dynamic pressure, an air data parameter that can be measured (or otherwise synthesized) using the aerodynamic air data system. Verification can be based on the consistency of all these data, although there may be some correlation between these parameters.

[0100] In some cases, multiple ways of estimating a given parameter may exist. Estimator function 138 can integrate various estimators for any single air data parameter, where each estimator is based on a different model / algorithm and a set of input parameters. In some embodiments, any desired air data parameter can be synthesized using different types of estimators and / or sets of input signals, thus generating multiple instances of parameter values. In some embodiments, each instance of a parameter value may include an evaluation of its accuracy based on the known corresponding accuracy of one or more parameters used to compute that instance. Similarly, the values ​​of the preliminary synthesized data 134 generated by estimator function 138 may include an evaluation of its accuracy based on the known corresponding accuracy of one or more parameters used to compute the values ​​of that preliminary synthesized data 134.

[0101] Figure 5B For those in training configuration Figure 5AA schematic diagram of the synthetic data system 135, in which the corrector function 140 is trained using ML in this training configuration. The corrector function 140 may not be based on a model dependent on prior knowledge of the aircraft 10 (e.g., physical), but rather on data obtained from a large training dataset via ML. The corrector function 140 may include a computer-implemented model trained using suitable ML training data (e.g., supervised ML) that associates previous (e.g., non-aerodynamic) additional data 32 with previously effective corrections 142B to be applied (e.g., at summing node 144) to the initial synthetic data 134 already generated by the estimator function 138. In some embodiments, the corrector function 140 may be implemented as one or more artificial neural networks (ANNs) of suitable type and architecture, and / or other computer-implemented models trained using ML. In various embodiments, the corrector function 140 may be implemented as a feedforward neural network, a recurrent neural network, and / or a radial basis function. An ANN in the form of a recurrent neural network can be suitable for estimating parameters that have historical dependencies, such as airflow parameters, which require time to adjust and change. The ANN used for the corrector function 140 can have a single-layer or multi-layer topology.

[0102] In some embodiments, a single, individual corrector function 140 may be associated with each synthesized parameter. Alternatively, a multidimensional corrector function 140 may generate multiple synthesized parameters. In some embodiments, the synthetic data system 135 may include combinations of one or more individual corrector functions 140 and one or more multidimensional corrector functions 140, whose inputs are the same, similar, or different.

[0103] The corrector function 140 may include: a single (i.e., unique) ANN for computing preliminary corrections 142A for a plurality of operating parameters of the aircraft 10; an ANN for computing preliminary corrections 142A for a corresponding operating parameter among the plurality of operating parameters of the aircraft 10; and / or a combination of ANNs, each ANN computing preliminary corrections 142A for one or more operating parameters of the aircraft 10. Furthermore, multiple ANNs driven by different sets of input signals may exist to generate preliminary corrections 142A to be applied to any given set of values ​​of preliminary synthetic data 134 associated with different operating parameters of the aircraft 10. In embodiments where the corrector function 140 comprises multiple independent ML training models with potentially different architectures, different applicable training procedures may be used for each machine learning training model. Training does not need to be performed simultaneously on all ML training models.

[0104] The ANN of the corrector function 140 can be trained using training algorithm 82 to minimize the loss function by optimizing one or more ANN parameters 84 to minimize the difference between the predicted output and the actual target value in a given dataset. Training algorithm 82 can implement gradient-based methods (e.g., backpropagation) to estimate the ANN parameters 84. During the training phase, the ANN can learn from suitable training data by iteratively updating the ANN parameters 84 to minimize the defined loss function. This method allows the ANN to generalize to unseen data. The ANN parameters 84 can represent connection weights and neuron (i.e., node) biases or thresholds. The ANN architecture (e.g., number of layers, number of neurons per layer, connection topology including feedforward and recursive types) can be selected to provide accurate correction across the entire operating domain while minimizing the required computation.

[0105] Training of the corrector function 140 can be performed using a training algorithm running on computer 28, FCC 28B, or other computers. For example, training of the corrector function 140 can be performed on a general-purpose central processing unit, GPU, NPU, or dedicated ANN software and / or hardware (such as the neural processing unit of the trade name Apple Neural Engine (ANE)).

[0106] In some embodiments, the corrections learned and generalized by the ANN can be transformed and implemented as one or more (e.g., multidimensional) lookup tables (LUTs). One or more LUTs can be extracted from a trained ANN by driving the ANN with a combination of input parameter values ​​that cover the full range of values ​​and whose resolution (i.e., discretization) matches the obtained and desired accuracy of the correction. The resulting LUTs can be validated to ensure that any step discontinuities in the correction data are correctly captured. For example, validation can be performed by extracting one or more LUTs at various discretization levels and comparing the computed corrections between them.

[0107] The training data provided to the training algorithm 82 for determining the ANN parameters 84 may include other data 32 and corresponding required corrections 86. The initial synthetic data 134 can be determined based on the other data 32 using the estimator function 138. The required corrections 86 can be determined by comparing the initial synthetic data 134 with the true data 88 at the summation node 90. The required corrections 86 may include the magnitude difference between the values ​​of the initial synthetic data 134 and the true data 88. In other words, the required corrections 86 can be determined by subtracting the initial synthetic data 134 from the true data 88 at the summation node 90. To maintain independence, the input to the correction function 140 may not include data subsequently compared with the initial synthetic data 134.

[0108] Real data 88 can represent valid primary data 24. However, real data 88 can also be determined through experience or through simulation / modeling. For example, real data 88 can be generated based on computational fluid dynamics, flight dynamics simulators (steady-state and unsteady-state flight and maneuvers), wind tunnel tests, recorded flight test data, or any combination thereof to improve the accuracy and robustness (i.e., generalization) of real data 88. Training data can include steady-state conditions and / or trim conditions. For example, training data can cover one or more of the following conditions: level flight; dynamic maneuvers such as turns and transitions; extreme conditions (such as near stall, high-speed dive); various airspeeds or Mach numbers; different turbulence levels; different atmospheric conditions and altitudes; a range of aircraft weights and center of gravity positions; different normal and abnormal (such as malfunction) aircraft configurations (such as retraction or deployment of lift-enhancing surfaces, retraction or deployment of landing gear, or during transitions, single-engine inactivity (OEI), etc.). The training data can cover conditions within the operational envelope of the aircraft 10, and can also cover certain anomalous conditions slightly outside the operational envelope, in order to improve accuracy and desired behavior at the boundaries of the operational envelope and slightly beyond the operational envelope.

[0109] In some embodiments, the training objective may be to enable the corrector function 140 to determine a correction that minimizes the estimation error between the corrected synthetic data 136 and the corresponding ground truth data 88 when applied to the initial synthetic data 134. In other words, the ANN parameter 84 of the corrector function 140 can be adjusted through training so that the initial correction 142A corresponds as closely as possible to the difference between the initial synthetic data 134 and the ground truth data 88. For example, the training may be designed to predict the error between the initial synthetic data 134 and the ground truth data 88, or to generate a correction that minimizes the error between the corrected synthetic data 136 and the corresponding ground truth data 88.

[0110] The training of the corrector function 140 can be performed offline, i.e., not during the flight of the aircraft 10. Therefore, the ANN parameters 84 can be predetermined, and the corrector function 140 can be pre-trained before being used as part of a real (e.g., aircraft) application.

[0111] Some embodiments of the synthetic data system 135 may allow online training of the corrector function 140 while the aircraft 10 (or flight test aircraft) is in operation and in flight. During online training, primary data 24, which is positively evaluated as valid, correct, and accurate, can be used as true data 88 to determine ANN parameters 84. Online training can be used to improve (fine-tune) the quality of the initial correction 142A using additional real-time data, thereby modifying the pre-trained correction function 140. During online training of the corrector function 140, the use of the corrected synthetic data 136 by the ISM65 and / or consumption system 26 can be (e.g., temporarily) disabled. In some embodiments, the synthetic data system 135 may include an offline-trained corrector function and an additional online-trained corrector function added to the offline-trained corrector. In such embodiments, the corrections determined by the online-trained corrector function may be limited. Training of the corrector function may be preferred for certification purposes. Some online training may be performed during flight of the flight test aircraft.

[0112] Figure 6 This is a schematic diagram of an exemplary architecture of an ANN for the corrector functions 40 and 140 of the synthetic data systems 35 and 135. The ANN may include interconnected units or nodes called artificial neurons. The circular nodes in the diagram represent artificial neurons, and the arrows each indicate a connection from the output of one neuron to the input of another. Each artificial neuron receives signals from connected neurons, processes them, and sends signals to other connected neurons. The signals may be real numbers, and the output of each neuron can be calculated as a function of the sum of its inputs (called an activation function). The strength of the signal at each connection is determined by weights (e.g., ANN parameter 84), which are iteratively adjusted during the learning process (i.e., training). The ANN can learn from experience and be trained using training data to derive a suitable correction 42 or preliminary correction 142A from other data 32.

[0113] The artificial neurons of this ANN can be aggregated into layers. Different layers can perform different transformations on their inputs. Signals can be passed from the input layer (e.g., the values ​​of other data 32) through one or more intermediate hidden layers 92 to the output layer (e.g., the corresponding values ​​of correction 42 or preliminary correction 142A). In some embodiments, the ANN can be a deep neural network with two or more hidden layers 92. In other words, the ANN can be used to associate one or more values ​​of other data 32 with the corresponding values ​​of correction 42 or preliminary correction 142A during the operation of corrector functions 40, 140. In a recursive ANN, in addition to the signals that travel from the input layer to the output layer through intermediate hidden layers (if any), some signals can also travel from the output layer to the input layer. In a recursive ANN, some signals can also connect downstream layers to upstream layers, regardless of whether these layers are hidden layers. For example, some signals can connect neurons in any intermediate layer in the direction toward the input layer (i.e., the reverse direction). In a recursive ANN, the output signal from a specific neuron in a layer can be fed back to itself as an input signal with a delay.

[0114] The recursive ANN topology can be applied to learn the dynamic characteristics (i.e., time dependencies) of correction 42 or preliminary correction 142A. For example, the current correction may depend on past aircraft states, not just the instantaneous state of aircraft 10. For instance, during a large maneuver, the airflow cannot adjust to the new state instantaneously, and therefore the correction may be slightly delayed. In some cases, another way to implement the time dependency aspect of correction 42 or preliminary correction 142A is to drive the corrector functions 40, 140 with a series of other data 32, not just the current other data 32, in any execution cycle. For example, corrector functions 40, 140 can be driven by multiple (e.g., sequential) previous values ​​(i.e., the last N instances) of other data 32, not just the current (latest) value of other data 32. The training of corrector functions 40, 140 can be coordinated with such sequential inputs of other data 32.

[0115] Figure 7 This is a flowchart of a method 1000 for monitoring the status of aircraft 10 or another aircraft. Method 1000 can be performed using synthetic data systems 35, 135 and / or another system. For example, machine-readable instructions 52 can be configured to cause computer 28 (e.g., FCC 28B) to perform at least a portion of method 1000. Method 1000 may include other actions disclosed herein. Method 1000 can be used within FCS 46. Method 1000 may include elements of monitoring system 12, FCS 46 and / or synthetic data systems 35, 135. In various embodiments, method 1000 may include: Use a first sensor (e.g., main sensor 22) to acquire key values ​​(e.g., key data 24) that indicate the aircraft's status (box 1002). The preliminary composite value indicating the aircraft's state is calculated using other data 32 acquired via a second sensor (e.g., other data source 30), which operates differently from the first sensor (box 1004). A computer-implemented model (e.g., corrector functions 40, 140) is used to compute correction values ​​(e.g., corrections 42, 142B) to be applied to the initial synthesized values. This computer-implemented model has been trained using ML and training data that correlates previous values ​​of other data 32 with previous correction values ​​(box 1006); and The correction values ​​are applied to the initial composite values ​​to generate corrected composite values ​​that indicate the aircraft's state (e.g., corrected composite data 36, ​​136) (Box 1008).

[0116] The primary value can be acquired when the first sensor (e.g., main sensor 22) interacts with an airflow outside the aircraft 10. The first sensor may be an aerodynamic sensor.

[0117] The computer-implemented model (e.g., corrector functions 40, 140) may include an ANN. Method 1000 may include training the ANN using previously corrected values ​​associated with previous values ​​of other data 32 before generating initial synthetic values.

[0118] Method 1000 may include: validating the principal value; and after generating the initial synthesized value, further training the ANN using ML and a new correction value based on the difference between the principal value and the initial synthesized value.

[0119] Method 1000 may include, before applying the correction value to the preliminary synthesized value: determining that the correction value exceeds a predetermined value; and adjusting the correction value to the predetermined value (e.g., using a limiter 176).

[0120] Calculating the correction value may include: identifying the current operating point of the aircraft 10 within the operating envelope of the aircraft 10; and calculating the correction value based on the current operating point of the aircraft 10 within the operating envelope.

[0121] The aircraft state may include one or more of the following: the airspeed of the aircraft 10; the Mach number of the aircraft 10 during travel; the angle of attack of the aircraft 10; the sideslip angle of the aircraft 10; the dynamic pressure of the air through which the aircraft 10 travels; the temperature of the air through which the aircraft 10 travels; the pressure altitude of the aircraft 10; and the vertical speed of the aircraft 10.

[0122] Figure 8 This is a flowchart of a method 2000 for operating aircraft 10 or another aircraft. Method 2000 may include some or all of method 1000. Method 2000 may be performed using synthetic data systems 35, 135, FCS 46 and / or another system. For example, machine-readable instructions 52 may be configured to cause computer 28 (e.g., FCC 28B) to perform at least a portion of method 2000. Method 2000 may include other actions disclosed herein. Method 2000 may include elements of monitoring system 12, FCS 46 and / or synthetic data systems 35, 135. In various embodiments, method 2000 may include: During flight of the aircraft 10, the main sensor 22, which interacts with the airflow outside the aircraft 10, acquires main aerodynamic data (e.g., main data 24), which indicates the aircraft's state (box 2002). Using non-aerodynamic data from other data sources 30 besides the main sensor 22, preliminary synthetic data 34, 134 (box 2004) indicating the state of the aircraft are generated. A computer-implemented model (e.g., corrector functions 40, 140) is used to generate corrections 42, 142B to be applied to the initial synthetic data 34, 134. This computer-implemented model has been trained with ML and training data that correlates previous non-aerodynamic data with previous corrections (box 2006). Corrections 42 and 142B are applied to the initial synthetic data 34 and 134 to generate corrected synthetic data 36 and 136 (box 2008) that indicate the state of the aircraft. It was determined that the primary aerodynamic data was unreliable or unavailable (box 2010); and After determining that the primary aerodynamic data is unreliable or unavailable, actions are performed on the aircraft 10 based on the corrected synthetic data 36, ​​136 (box 2012).

[0123] In some embodiments, the action may include controlling one or more flight control surfaces 20 of the aircraft 10 automatically via FCS 46 or via manual input by the crew. In some embodiments, the action may include adjusting actuators on the aircraft 10. In some embodiments, the action may include transmitting corrected synthetic data 36, ​​136 to the crew via cockpit displays or other warning devices.

[0124] Method 2000 may include: controlling the flight control surface 20 of the aircraft 10 based on the primary aerodynamic data before determining that the primary aerodynamic data is unreliable or unavailable.

[0125] Method 2000 may include: when it is determined that the primary aerodynamic data is unreliable or unavailable, comparing the primary aerodynamic data with the corrected synthetic data 36, ​​136.

[0126] In some embodiments of method 2000, applying corrections 42, 142B to preliminary synthetic data 34, 134 may include adding correction values ​​to the synthetic values ​​of the preliminary synthetic data 34, 134.

[0127] In some embodiments of method 2000, applying corrections 42, 142B to the preliminary synthesized data 34, 134 may include multiplying the correction value with the synthesized value of the preliminary synthesized data 34, 134.

[0128] Method 2000 may include, before applying corrections 42, 142B to preliminary composite data 34, 134: determining that the correction value exceeds a predetermined value; and reducing the correction value (e.g., using a limiter 176).

[0129] Generating corrections 42 and 142B may include: using preliminary synthetic data 34 and 134 to identify the operating point of the aircraft 10 within the operating envelope of the aircraft 10; and calculating corrections 42 and 142B based on the operating point within the operating envelope.

[0130] The aircraft state may include one or more of the following: the airspeed of the aircraft 10; the Mach number of the aircraft 10 during travel; the angle of attack of the aircraft 10; the sideslip angle of the aircraft 10; the dynamic pressure of the air through which the aircraft 10 travels; the temperature of the air through which the aircraft 10 travels; the pressure altitude of the aircraft 10; and the vertical speed of the aircraft 10.

[0131] Aspects of this disclosure may be embodied in systems, apparatus, methods, and / or computer program products for estimating the remaining useful life of components of an aircraft engine or for performing other methods described herein. For example, aspects of this disclosure may take the form of a computer program product embodied in one or more non-transitory computer-readable media (e.g., memory 50), on which computer-readable program code (e.g., instructions 52, corrector functions 40, 140, estimator functions 38, 138, and training algorithm 82) is contained. The program code may be read and executed by one or more computers (e.g., computer 28, FCC 28B), processors, or logic circuits to perform any of method 1000 and method 2000.

[0132] Figure 9 This is a table illustrating an exemplary data structure 94, exemplifying the data used to generate the ML training dataset 96 (shown in...). Figure 10 The method. Figure 10This is a table showing an exemplary ML training dataset 96 that can be provided to training algorithm 82. Also refer to... Figure 5B The initial synthetic data 134 (e.g., values ​​S1-S3) can be determined by the estimator function 138 based on other data 32 (e.g., values ​​O1-O3). The required correction 86 (e.g., values ​​C1-C3) can include the magnitude difference between the initial synthetic data 134 (e.g., values ​​S1-S3) and the true data 88 (e.g., values ​​T1-T3), for example, making C1 = T1 - S1. The ML training dataset 96 can include previous (i.e., previously determined) values ​​of other data 32 (e.g., O1, O2, O3, ...) and the corresponding previous (i.e., previously determined) values ​​of the required correction 86 (e.g., C1, C2, C3, ...), which are to be provided as ML training data to the training algorithm 82, thereby generating ANN parameters 84 and training the corrector function 140 (e.g., the ANN of the corrector function 140) in an iterative and supervised ML manner.

[0133] Therefore, as will be clear, the above and illustrated examples are intended to be exemplary only. The scope is defined by the appended claims.

Claims

1. A method for operating an aircraft, the method comprising: During the flight of the aircraft, key aerodynamic data are acquired using sensors that interact with airflow outside the aircraft, and the key aerodynamic data indicates the aircraft's status. Using non-aerodynamic data from sources other than the aforementioned sensors, preliminary synthetic data indicating the state of the aircraft is generated; The computer-implemented model is used to generate a correction for the initial synthetic data. The computer-implemented model has been trained using machine learning and training data that correlates previous non-aerodynamic data with previous corrections. The correction is applied to the initial synthetic data to generate corrected synthetic data that indicates the state of the aircraft; It was determined that the key aerodynamic data was unreliable or unavailable; and After determining that the primary aerodynamic data is unreliable or unavailable, actions are performed on the aircraft based on the corrected synthetic data.

2. The method of claim 1, wherein the action includes controlling the flight control surface of the aircraft.

3. The method according to claim 2, comprising: The flight control surfaces of the aircraft are controlled based on the primary aerodynamic data until it is determined that the primary aerodynamic data is unreliable or unavailable.

4. The method according to any one of claims 1 to 3, comprising: When it is determined that the primary aerodynamic data is unreliable or unavailable, the primary aerodynamic data is compared with the corrected synthetic data.

5. The method according to any one of claims 1 to 4, wherein the action comprises: The corrected synthetic data is transmitted to the crew of the aircraft.

6. The method according to any one of claims 1 to 5, wherein applying the correction to the preliminary synthetic data comprises adding the correction value to the synthetic value of the preliminary synthetic data.

7. The method according to any one of claims 1 to 5, wherein applying the correction to the preliminary synthesized data comprises multiplying the correction value by the synthesized value of the preliminary synthesized data.

8. The method of claim 6 or claim 7, comprising, before applying the correction to the preliminary synthetic data: It is determined that the correction value exceeds the specified value; and Adjust the correction value to the specified value.

9. The method according to any one of claims 1 to 8, wherein generating the correction comprises: The preliminary synthetic data is used to identify the operational points of the aircraft within the operational envelope of the aircraft. as well as The correction is calculated based on the operation point within the operation envelope.

10. The method according to any one of claims 1 to 9, wherein the aircraft state includes the airspeed of the aircraft.

11. The method according to any one of claims 1 to 9, wherein the aircraft state includes the Mach number of the aircraft's travel.

12. The method according to any one of claims 1 to 9, wherein the aircraft state includes the angle of attack of the aircraft.

13. The method according to any one of claims 1 to 9, wherein the aircraft state includes the sideslip angle of the aircraft.

14. The method according to any one of claims 1 to 9, wherein the aircraft state includes the dynamic pressure of the air through which the aircraft travels.

15. The method according to any one of claims 1 to 9, wherein the aircraft state includes the temperature of the air through which the aircraft travels.

16. The method according to any one of claims 1 to 9, wherein the aircraft state includes the pressure altitude of the aircraft.

17. The method according to any one of claims 1 to 9, wherein the aircraft state includes the vertical velocity of the aircraft.

18. The method according to any one of claims 1 to 9, wherein the aircraft state includes the air density outside the aircraft.

19. The method according to any one of claims 1 to 9, wherein the aircraft state includes the weight of the aircraft.

20. A method for monitoring the status of an aircraft, the method comprising: The first sensor is used to acquire key values ​​indicating the state of the aircraft; A preliminary composite value indicating the state of the aircraft is calculated using other data acquired via a second sensor, the second sensor operating differently from the first sensor; A computer-implemented model is used to calculate the correction value to be applied to the initial synthesized value. The computer-implemented model has been trained with machine learning and training data that correlates previous values ​​of the other data with previous correction values. as well as The correction value is applied to the initial composite value to generate a corrected composite value that indicates the state of the aircraft.

21. The method of claim 20, wherein the principal value is acquired when the first sensor interacts with an airflow outside the aircraft.

22. The method according to claim 20 or 21, wherein: The computer-implemented model includes artificial neural networks; as well as The method includes training the artificial neural network using previous values ​​of the other data associated with the previous correction values ​​before generating the initial synthesized values.

23. The method of claim 22, comprising: Verify the stated primary value; as well as After generating the initial synthesized value, the artificial neural network is further trained using machine learning and new correction values ​​based on the difference between the primary value and the initial synthesized value.

24. The method according to any one of claims 20 to 23, comprising, before applying the correction value to the preliminary synthesized value: It is determined that the correction value exceeds the specified value; and Decrease the correction value.

25. The method according to any one of claims 20 to 24, wherein calculating the correction value comprises: Identify the operational points of the aircraft within the operational envelope of the aircraft; as well as The correction value is calculated based on the aircraft's operating point within the operating envelope.

26. The method according to any one of claims 20 to 25, wherein the aircraft state includes one or more of the following: the airspeed of the aircraft; the Mach number of the aircraft; the angle of attack of the aircraft; the sideslip angle of the aircraft; the dynamic pressure of the air through which the aircraft travels; the temperature of the air through which the aircraft travels; the pressure altitude of the aircraft; the air density; the weight of the aircraft; and the vertical velocity of the aircraft.

27. The method according to any one of claims 20 to 26, wherein the other data acquired via the second sensor includes a plurality of values ​​acquired at different times.

28. A computer program product for operating an aircraft, the computer program product comprising a non-transitory computer-readable storage medium containing program code that can be read / executed by a computer, processor, or logic circuitry to perform the method of any one of claims 1 to 27.

29. An aircraft system comprising: Sensors that are capable of operating to acquire key data indicating the status of the aircraft; One or more data processors; as well as A non-transitory machine-readable storage device, wherein the non-transitory machine-readable storage device stores: An estimator function, which is used to generate preliminary synthetic data from data acquired from sources other than the sensor; A corrector function, wherein the corrector function is used to generate corrections to be applied to the initial synthetic data using a computer-implemented model, the computer-implemented model being trained using machine learning and training data that correlates previous corrections with other previous data; and Instructions, which can be executed by the one or more data processors and are configured to cause the one or more data processors to perform the following operations: The estimator function is used to generate preliminary synthetic data indicating the state of the aircraft using the other data; The corrector function is used to generate corrections to be applied to the initial synthesized data; The correction is applied to the initial synthetic data to generate corrected synthetic data that indicates the state of the aircraft; as well as Generate an output indicating the corrected synthetic data.

30. The aircraft system of claim 29, wherein the sensor is configured to be exposed to airflow outside the aircraft during operation of the sensor.

31. The aircraft system according to claim 29 or claim 30, wherein: The sensor is the first sensor; The system includes sources other than the sensor; and The source other than the aforementioned sensor includes a second sensor, the second sensor having a different operating principle than the first sensor.

32. The aircraft system according to any one of claims 29 to 31, wherein the instructions are configured to cause the one or more data processors to apply a limitation to the correction before applying the correction to the preliminary synthesized data.

33. The aircraft system of claim 32, wherein limitations on the correction application include: Determine that the magnitude of the correction exceeds a threshold; as well as Reduce the magnitude of the correction.

34. The aircraft system according to any one of claims 29 to 33, wherein the instructions are configured to cause the one or more data processors to apply a filter to the correction before applying the correction to the preliminary synthesized data.

35. The aircraft system according to any one of claims 29 to 34, wherein calculating the correction comprises: Identify the operational points of the aircraft within the operational envelope of the aircraft; as well as The correction is generated based on the aircraft's operating points within the operating envelope.

36. The aircraft system according to any one of claims 29 to 35, wherein the instructions are configured to cause the one or more data processors to perform the following operations: It was determined that the key data was unreliable or unavailable; and After determining that the primary data is unreliable or unavailable, action commands are issued on the aircraft based on the corrected synthetic data.

Images