JP2025528621A - Air data estimation in aircraft - Google Patents

Abstract

Air data estimation involves determining values of air data parameters such as airspeed, altitude, angle of attack, and sideslip angle. Accurate determination of air data parameter values is crucial to enable accurate control of the aircraft. The aircraft has an air data estimator that comprises a machine learning model trained to predict air data parameter values from input values. The air data estimator can operate in real time on input values, where the input values are sensor data from sensors internal to the aircraft only or are data derived from such sensor data.
[Selected Figure] Figure 1

Description

The present invention relates to aircraft, and in particular to air data estimation in such aircraft.

Air data estimation involves determining values for air data parameters such as airspeed, pressure altitude, angle of attack, and sideslip angle. Accurate determination of air data parameter values is crucial to enabling precise control of an aircraft. Air data parameter values are typically displayed to the pilot and also made available to avionics systems onboard the aircraft. In one example, the altitude hold capability of an autopilot system on an aircraft requires accurate altitude values as input. In one example, angle of attack and sideslip angle are used by an aircraft's flight control system to automatically improve the aircraft's stability and controllability.

According to one aspect of the present invention, there is provided an aircraft comprising:
an air data estimator comprising a machine learning model trained to predict air data parameter values from input values;
The air data estimator is operable in real time on input values, where the input values are sensor data from sensors internal to the aircraft only, or data derived from such sensor data.

Preferably, the aircraft includes a flight control system, and wherein the input values include values from the flight control system.

Preferably, the input value comprises a static pressure value.

Preferably, the inputs from the flight control system include control surface positions.

Preferably, the input values comprise engine data from a digital engine control unit.

Preferably, the aircraft is equipped with a fleet management system, and wherein the input value comprises the mass of the aircraft as determined by the fleet management system.

Preferably, the input value comprises the aircraft's center of gravity as determined by the fleet management system.

Preferably, the aircraft is equipped with a navigation subsystem and an inertial measurement unit that provide aircraft attitude data within input values.

Preferably, the navigation subsystem and inertial measurement unit provide a steering rate within the input values.

Preferably, the navigation subsystem and inertial measurement unit provide aircraft acceleration within the input values.

Preferably, the air data parameter values include any one or more of angle of attack, sideslip, dynamic pressure, and static pressure.

Preferably, the air data parameter values comprise static pressure and dynamic pressure, and the aircraft comprises an output parameter calculator configured to calculate an altitude of the aircraft and an airspeed of the aircraft from the static pressure and dynamic pressure parameter values predicted by the air data estimator.

Preferably, the machine learning model is a neural network having an input layer, an output layer, and up to two hidden layers.

Preferably, the neural network is a feedforward neural network without recursion.

Preferably, the neural network is implemented using hardware circuitry comprising a graphics processing unit or a parallel processing device.

Preferably, the machine learning model is trained using labeled training data computed by randomly selecting values from the aircraft's flight envelope and using the selected values to simulate input values.

According to one aspect of the present invention, there is provided a method carried out in an aircraft, the method comprising:
Using an air data estimator comprising a machine learning model for predicting air data parameter values from input values in real time, where the input values are sensor data from sensors internal to the aircraft only or data derived from such sensor data.

Preferably, the input values include engine data, control surface position, static pressure, mass, center of gravity, attitude, maneuver rate, and acceleration.

Preferably, the method comprises concatenating the input values into a vector and inputting the vector into a machine learning model.

Preferably, the method comprises normalizing the vectors before inputting them into the machine learning model.

Preferably, the method comprises denormalizing the output values output from the machine learning model.

In accordance with one aspect of the present invention, a computer-implemented air data estimator is provided that includes a machine learning model for predicting air data parameter values from input values in real time, where the input values are from avionics within the aircraft, such that the aircraft does not require air data sensors mounted on the aircraft's exterior.

Preferably, the air data estimator is implemented using hardware circuitry, and the machine learning model is a neural network with fewer than three hidden layers.

According to one aspect of the present invention, there is provided a computer-implemented method for training an air data estimator to predict values of air data parameters of an aircraft, the method comprising:
Using supervised training to train the neural network in the air data estimator, the supervised training uses training data pairs, each training data pair comprising ground truth values of air data parameters selected from the flight envelope of the aircraft and corresponding simulated values from avionics in the aircraft, the simulated values being calculated using stored empirical data for the aircraft's engines and empirical data for the aircraft obtained from wind tunnel testing, and rules of computational fluid dynamics.

Preferably, the neural network comprises multiple neural networks, one for each parameter to be predicted.

Embodiments of the present invention will now be described, by way of example only, with reference to the drawings, in which:

FIG. 1 shows an aircraft and has a schematic diagram of various avionics within the aircraft, including an air data estimator for estimating air data parameter values. FIG. 2 shows an aircraft subjected to aerodynamic forces, inertial forces and gravity forces. FIG. 3 shows an exemplary method for calculating training data. FIG. 4 shows an example of an air data estimator. FIG. 5 shows an example of a neural network used in the air data estimator. FIG. 6a shows a flow diagram of an exemplary method for training a neural network for use in an air data estimator. FIG. 6b shows a flow diagram of an exemplary method for validating a neural network for use in an air data estimator. FIG. 7 shows a flow diagram of an exemplary method for using an air data estimator in an aircraft. FIG. 8a is a graph of the sideslip angle estimated by the air data estimator versus the ground truth sideslip angle. FIG. 8b is a graph of the equivalent airspeed in knots estimated by the air data estimator and the ground truth equivalent airspeed in knots. FIG. 8c is a graph of the angle of attack estimated by the air data estimator and the ground truth angle of attack.

Traditional air data systems on aircraft typically use data from surface-mounted sensors that measure barometric pressure and air temperature, from which airspeed, altitude, and ambient temperature are derived. Airflow detectors may also be mounted on aircraft as stall warning devices. More sophisticated aircraft may also require sensors that measure airflow angles, from which angle of attack and sideslip angle can be determined for use by the pilot as well as the aircraft's flight control system to improve the aircraft's stability and handling.

Conventional air data probes are mounted on the aircraft's forebody, a protruding object that is vulnerable to damage both on the ground and in the air and can become clogged with debris. External air data probes must also be heated to prevent ice buildup or water accumulation. These sensors also create aerodynamic drag and can cause aerodynamic interference with other surface-mounted protrusions, potentially even degrading engine intake performance. Externally mounted probes can also significantly contribute to the aircraft's radar signature. On high-performance military aircraft, optimal external locations for air data sensors are often not possible; i.e., radar performance requirements may prevent mounting sensors on the radome, flight and mission sensors may compete for location on the forebody, and the location of landing gear, engine air inlets, and control surfaces may also limit air data sensor locations. Furthermore, modern high-performance military aircraft mount multiple sensors on the forebody, and integrating multiple air data sensors with them is often difficult. Therefore, the ability to determine air data without the use of external air data probes is highly desirable.

The inventors have developed a method for determining various air data parameters without using data from conventional air data sensors. Only sensor data from, or derived from, sensors internal to the aircraft is used. Machine learning models are used to predict air data parameter values in real time, i.e., while the aircraft is in flight, so that the predicted air data parameter values can be used to control the aircraft during its flight. The technique is operable on any aircraft (including propeller-driven aircraft) equipped with suitable inertial and flight control hardware, as described below. The air data estimator described herein is operable for both civilian and military aircraft. While there are advantages to implementing air data estimation in both applications, there are likely greater advantages in the case of military aircraft, where air data sensor installation is more challenging and the flight and maneuvering envelope is larger.

Air data requires both accuracy and redundancy. Having an air data estimator as described herein provides another source of air data that is largely independent of other air data sources.

Developing machine learning models operable to predict air data parameter values in real time for use in aircraft is not easy. This problem is particularly exacerbated in the case of military aircraft, or fighter jets. Jet fighter aircraft are highly agile and are used to perform complex maneuvers in which the values of air data parameters change rapidly and move between extremes. Fighter jets operate in a range of more severe environmental conditions than commercial aircraft and experience greater turbulence than typical commercial aircraft.

Jet fighter aircraft are highly resource-constrained devices. Space, power, weight, and time are examples of scarce resources in the case of jet fighter aircraft. Therefore, deploying machine learning models for air data estimation in jet fighter aircraft faces the challenge of constrained resources.

FIG. 1 illustrates an aircraft 100 and includes a schematic diagram of various avionics within the aircraft, including an air data estimator 106 for estimating values of air data parameters 120. The air data estimator 106 and output parameter calculator 112 within the aircraft 100 output values of the air data parameters 120. The air data estimator 106 and output parameter calculator 112 are computer-implemented and execute an air data process 105. In the example of FIG. 1, the air data parameters 120 comprise angle of attack 122, sideslip 124, pressure altitude 126, and airspeed 128. The air data estimator 106 and output parameter calculator 112 are capable of calculating values of the air data parameters 120 in real time without using air data sensors on the exterior of the aircraft 100. This is accomplished by using avionics within the aircraft to provide values for input to a machine learning model in the air data estimator 106. The avionics may be the aircraft's existing avionics; that is, the avionics do not need to be modified for use with the air data estimator 106. The air data estimator 106 is computer-implemented using any one or more of software, firmware, and hardware. In one example, the air data estimator 106 is a low-power, compact computer-implemented component that is easy to deploy inside the aircraft.

Angle of attack 122 is the angle between the aircraft's baseline and the oncoming flow.

Slip, also known as sideslip angle, is the angle between the direction an aircraft is heading and the direction of the oncoming air.

Pressure altitude is a function of static pressure according to the agreed-upon definition of the International Standard Atmosphere. Therefore, pressure altitude can be derived from the corrected static pressure from the air data estimator.

Airspeed can be either knot equivalent airspeed, which is a calibrated airspeed adjusted for compressibility effects, or calibrated airspeed. Calibrated airspeed (CAS) is a function of pitot minus static pressure and is the airspeed typically displayed to the pilot.

Equivalent airspeed (EAS) is CAS corrected for the effects of air compressibility and is a function of dynamic pressure. Mach number is the ratio of true airspeed to the local speed of sound and can be calculated from pitot and static pressures. It can also be derived from static and dynamic pressures. When total temperature data is measured by a probe installed in the engine intake duct and used in the airspeed calculation, this airspeed is the true airspeed. In some cases where a speed ratio is calculated, the Mach number is given.

The values of the air data parameters 120 calculated using the air data estimator 106 and the output parameter calculator 112 are optionally displayed on a display 118 on the aircraft 100.

As mentioned above, the avionics may be existing avionics within the aircraft. The example avionics of FIG. 1 includes a digital engine control unit 108, a fleet management system 110, a flight control system 114, a navigation subsystem, and an inertial measurement unit 116. Note that one or more of the avionics of FIG. 1 may be integrated together in some examples. In one example, the flight control system 114 and the air data estimator 106 are integrated together.

Digital engine control units 108 are avionics that monitor and automatically adjust aircraft engine performance and related metrics throughout flight. They perform electronic management of the engines for optimized engine performance and perform engine health monitoring. Digital engine control units 108 receive data from sensors on the aircraft engines. Each engine has electronic controls mounted on the engine or engine fan case that draw power from the engine alternator to receive engine data from sensors on the engine and engine systems that measure engine data such as one or more of vibration, fuel consumption, thrust, power usage, temperature, emissions, and sound.

The vehicle management system 110 is an avionics system that includes a computer system that receives data from sensors on the aircraft, which monitors the amount of fuel on the aircraft and the number of weapons on the aircraft. The vehicle management system outputs the current mass of the aircraft and the current center of gravity of the aircraft. The mass and center of gravity of the aircraft are quantities that change as fuel is burned (or accepted in the case of in-flight refueling) and weapons are released. In some cases, the vehicle management system 110 is part of the flight control system 114. More details about the flight control system are provided next. The vehicle management system 110 can, in some examples, perform many other functions.

The flight control system 114 is an avionics system that operates the aircraft's control surfaces. The flight control system 114 operates the control surfaces either automatically (e.g., by issuing computer-implemented commands to stabilize the aircraft without pilot input) or with input from a human pilot operating cockpit controls in the aircraft. Control surfaces, such as ailerons, elevators, rudders, spoilers, flaps, slats, air brakes, and control trimming surfaces, are aerodynamic devices that can move to control the aircraft's flight attitude. In one example, hinges or tracks are used to enable the movement of the control surfaces to deflect the airflow passing over the control surfaces and cause the aircraft to rotate about an axis associated with the control surfaces. The flight control system has information regarding the positions of the aircraft's control surfaces and can provide the control surface positions as values for input to the air data estimator 106. The flight control system may also have multiple air data sensors, including one or more static pressure sensors integrated into dedicated sensor/transducer units. Thus, static pressure may be provided to the air data estimator regardless of whether the air data estimator is implemented in a flight control computer or a dedicated air data computer. In one example, an aircraft's static pressure system includes static ports, which are small openings in the aircraft's hull that allow sensing of ambient atmospheric pressure during flight. The flight control system may provide measured static pressure values as inputs to the air data estimator 106. In some cases, static pressure is provided directly to the air data estimator or flight control system from a dedicated pressure sensor unit. Static pressure data, in some examples, is digitized and provided via a digital data bus.

The navigation subsystem and inertial measurement unit 116 comprises one or more accelerometers, gyroscopes, or global positioning systems within the aircraft. The navigation subsystem and inertial measurement unit 116 are avionics and sensors within the aircraft that output one or more of the following: aircraft attitude, such as the aircraft's orientation relative to the horizon; aircraft maneuver rate; and aircraft acceleration. The aircraft maneuver rate is the amount of maneuvers executed per unit time. Considering a basic maneuver such as climbing altitude, here the rate of maneuver is the rate of change of altitude. Other types of maneuvers are turns, which result in a change in the aircraft's heading, and accelerations, which result in a change in the aircraft's speed and Mach number. Complex maneuvers involve combinations of more than one type of basic maneuver.

The values of the air data parameters 120 calculated by the air data estimator 106 and the output parameter calculator 112 are displayed on a display 118 in the aircraft 100 and/or used by avionics in the aircraft 100, such as the flight control system 114, the digital engine control unit 108, the fleet management system 110, and the navigation subsystem and inertial measurement unit 118.

Figure 2 shows an aircraft subjected to aerodynamic, inertial, and gravitational forces during flight (L represents lift, W represents gravity, and D represents drag). Also shown in Figure 2 are the angle of attack alpha and sideslip angle beta, and the airflow velocity vector V. The aircraft produces engine thrust T as shown. When these forces are balanced, the aircraft is said to be "trimmed." While engine thrust and control surface position are constant in the trimmed condition, they depend on the aircraft's mass and center of gravity, which change as fuel is burned (or accepted in the case of in-flight refueling) and weapons are released.

The physics of aircraft flight are embodied in equations of motion that take into account aerodynamic forces, inertial forces, and gravity. Using data from the avionics described in Figure 1, information about inertial and gravitational forces is known. Aerodynamic forces are a function of airspeed and onset airflow direction, the latter defined by angle of attack and sideslip angle. Aerodynamic forces are also affected by any configuration changes that affect the aircraft's external line, such as undercarriage deployment. Whether undercarriage deployment is activated is known to the vehicle management system in the avionics of Figure 1.

Figure 3 shows an example method for calculating training data for training the machine learning model in the air data estimator 106. In the example of Figure 3, the machine learning model is trained using supervised training, although unsupervised or semi-supervised training can also be used. Accuracy is measured quantitatively using validation data, as described below, and takes into account generalization ability. Generalization ability is assessed by ensuring that the validation data points are sufficiently distant from the training set data points. This is accomplished by dividing the dataset into batches/runs (approximately 3000 points) rather than pure random sampling. In some cases, accuracy is also assessed qualitatively using feedback from human pilots. Generalization ability is a measure of how well the machine learning model can predict values of air data parameters from input values that were not represented in the training data.

In supervised training, labeled training data pairs are used. A training data pair is a set of values from the avionics on an aircraft at a specified time and a corresponding set of ground truth air data parameter values. The ground truth air data parameter values are known to be correct at the specified time. Hundreds of thousands of training data pairs are used during training to achieve high accuracy for the machine learning model in the air data estimator. The training data pairs are obtained for a wide range of possible aircraft states, including different attitudes, air data speeds, environmental conditions such as turbulence, and different internal conditions such as mass, engine state, center of gravity, landing gear deployment, and control surface position. Using a wide range of possible aircraft states can improve the accuracy and generalization ability of the machine learning model.

Obtaining a sufficient amount and sufficient diversity of training data pairs for training a machine learning model is not easy. One option is to fly an aircraft and record a set of values from the avionics on board the aircraft and a corresponding set of empirical ground truth air data parameter values. In one example, the empirical ground truth air data parameter values are measured using a flight test boom instrumented with air data sensors. However, because extremes in environmental conditions and/or extremes in attitude and airspeed are difficult to achieve, obtaining sufficiently diverse training data pairs by flying an aircraft and recording data is extremely difficult.

The inventors have developed a method for generating large volumes and diverse training data pairs through the use of simulation, where the simulation is a computer simulation that uses an integrated flight control model of aircraft dynamics and aerodynamics, engine performance, and flight control systems. In one example, characteristics are measured in a wind tunnel or generated by a computational fluid dynamics solver. The resulting simulator is then used to generate training data.

Training data is generated to cover the input space 300 of the air data estimator. The input space is the range of all input parameters of the air data estimator.

As shown in FIG. 3, values from one or more of the aircraft's avionics components are simulated for each of a range of values for the air data parameters. Thus, for a particular set of simulated values 316 for the aircraft's avionics components, there is a corresponding set of ground truth values 318 for the air data parameters. By repeating the simulation process for many sets of ground truth values for the air data parameters, a large number of training data pairs 322 are obtained.

The flight envelope of the aircraft is known and stored in a database in the form of rules and/or restrictions. The flight envelope is useful information for at least a portion of the input space 300. The design envelope of the aircraft is typically provided by the aircraft manufacturer and includes Mach number, altitude, and airspeed, along with acceleration and maneuver rate restrictions. Ideally, the air data estimator is trained to function across the entire design envelope. The flight envelope in which the aircraft is cleared to fly is within the design envelope to provide a safety margin. The flight envelope is the range of altitudes, attitudes, speeds, and accelerations that can be achieved using the aircraft.

The computer-implemented selector 302, in some cases a random or pseudo-random selector, selects a set of values for the air data parameters that fall within the flight envelope 300. The computer-implemented selector 302 optionally takes into account constraints 304 such that a set of values for the input parameters is selected from the input space 300 in accordance with the constraints 304. An example of a constraint is a maximum acceleration (e.g., 9 G) that is practical for a human pilot to experience.

In some cases, the computer-implemented selector 302 takes maneuver trajectories into consideration when selecting a set of values for the air data parameters. Unlike conventional maneuvers, random inputs are provided frequently for random trajectories. Alternatively or additionally, one or more template maneuver trajectories are available to the random selector, where a template maneuver trajectory is a flight control input over time. Template maneuver trajectories are created for turns, altitude climbs, altitude declines, accelerations, subsonic, sonic, and supersonic speeds, banked turns, pitch ups, and combinations thereof. The computer-implemented selector 302 is configured to select multiple altitudes and attitudes along the trajectory and output corresponding ground truth air data parameters. The use of extreme random maneuvers can demonstrate the robustness of the air data estimator 106.

Once a set of air data parameter values is selected, corresponding values from the avionics in the aircraft are simulated. In one example, there is one or more of a digital engine control unit 108 simulation 306, a flight control system simulation 308, a fleet management system simulation 312, and a navigation subsystem and inertial measurement unit 116 simulation 314. One or more of these simulators may be combined into a single simulator in the flight control system simulator 308, rather than being separate entities as shown in FIG. 3.

In some examples, a random input generator is used to generate flight control values for a specified duration. In one example, altitude and speed are varied in a grid fashion to ensure coverage of those aspects.

In one example, the digital engine control unit simulation 306 comprises stored empirical data regarding engine performance determined from ground and flight test benches. One or more engines of the aircraft type being used are tested on the ground or in flight to obtain sensor data measurements as would be obtained by the digital engine control unit in use. The stored empirical data comprises fuel consumption, thrust, power usage, temperature, emissions, and sound. The digital engine control unit simulation 306 can take a set of air data parameter values as input and return engine data from the stored empirical data. This is done by using rules regarding how to query the stored empirical data according to the ground truth air data parameter values.

In one example, the vehicle management system simulation 312 includes software-implemented rules that output the current mass of the aircraft and the current center of gravity of the aircraft. The mass and center of gravity of the aircraft are quantities that change as fuel is burned (or accepted in the case of in-flight refueling) and weapons are released. The vehicle management system simulation 312 uses the rules to generate values by using default values, selecting values according to rules or models of fuel consumption by the aircraft over time, selecting values according to rules or models of weapons deployment by the aircraft over time, selecting values according to rules or models of in-flight refueling, etc.

In one example, the simulation 314 of the navigation subsystem and inertial measurement unit 116 comprises software that calculates one or more of the aircraft attitude, such as the orientation of the aircraft relative to the horizon, the aircraft maneuver rate, and the aircraft acceleration, given ground truth air data parameter values. The inertial data is generated in the simulation 314 using a set of template maneuvers for various conditions within the design envelope of the aircraft. The simulation 314 comprises one or more rules or lookup tables, such as for selecting an aircraft attitude from a range of values, given the ground truth air data parameters. The template maneuvers used by the selector 302 may also be used by the simulation 314 to inform the selection of the attitude.

Empirical aircraft data obtained from wind tunnel testing and/or computational fluid dynamics (CFD) rules are used to generate aerodynamic data sets and data for air data corrections that are embedded in the flight control model software.

In some cases, the simulation includes a software model of the inertial measurement unit that simulates measurements of the aircraft's acceleration, taking into account noise in the inertial measurement unit. In some cases, the simulation includes rules for calculating simulated maneuver rates for the aircraft given the ground truth air data parameters and the template maneuvers used by selector 302.

In one example, a flight control system simulation 308 comprises software that outputs aircraft control surface positions and static pressures given a set of maneuver demands and ground truth air data parameter values as inputs. The flight control system simulation 308 uses a flight control model to simulate aircraft movement from a trimmed condition through either template maneuvers or maneuvers with random but constrained pilot control inputs. The aircraft response is further constrained by the flight control system. In one example, static pressures are calculated by the simulation 308 according to altitude values provided in the ground truth parameter values and taking into account position errors and noise.

In some examples, turbulence is taken into account by the simulated flight control system and/or the simulated navigation subsystem and inertial measurement unit. In these examples, a wind gust model, such as the Dryden wind turbulence model, or other gust model that simulates both vertical and lateral wind gusts, is used by the simulator. In some examples, the turbulence is scaled to reflect maximum wind speeds (jet streams) and the resulting extreme turbulence. Randomly selected amounts of wind turbulence for both vertical and lateral wind gusts are simulated and taken into account by the simulator.

As shown in FIG. 3, simulated values 316 are calculated. One set of simulated values is a plurality of simulated values (representing aircraft conditions at a particular time or time interval) that are concatenated into a single vector and input to a machine learning model in the air data estimator. The process of FIG. 3 is repeated to simulate tens or hundreds of thousands of sets of simulated values. Each set of simulated values has a corresponding set of ground truth air data parameter values.

Bias correction 320 is optionally performed if there is an uneven distribution of outputs. An uneven distribution of outputs occurs when the neural network model is inherently biased to predict more frequent values. Bias correction through loss weights makes it possible to mitigate this.

The process of FIG. 3 generates data pairs, where a data pair is a set of simulated values from the avionics in an aircraft and a corresponding set of air data parameter values. A first portion of the data pairs generated using the process of FIG. 3 are stored in a training data store 322. A second portion of the data pairs generated using the process of FIG. 3 are stored in a validation data store 324. The validation data pairs differ from the training data pairs because they are taken from an entirely separate training run. There are many more data pairs in the training data store 322 than in the validation data store 324.

Figure 4 shows an example of an air data estimator 106, which is a machine learning model such as a neural network, a random decision forest, a support vector machine, or other machine learning model. In the example described herein, the machine learning model is a neural network, but one skilled in the art will understand that an equivalent machine learning model may be used in place of a neural network.

The air data estimator 106 receives inputs from one or more of a digital engine control unit (DECU) 108, a flight control system 114, a vehicle management system 110, and a navigation subsystem and inertial measurement unit 116. In one example, the DECU 108 sends engine data to the air data estimator. The flight control system 114 sends aircraft control surface positions and aircraft static pressures to the air data estimator. The vehicle management system 110 sends aircraft mass and center of gravity to the air data estimator 106. The navigation subsystem and inertial measurement unit sends one or more of aircraft attitude, maneuver rate, and aircraft acceleration to the air data estimator 106.

The air data estimator outputs a predicted angle of attack 122, a predicted sideslip angle 124, a predicted static pressure, and a predicted dynamic pressure. The predicted static pressure and the predicted dynamic pressure are input to an output parameter calculator 112. The output parameter calculator 112 calculates a predicted pressure 126 for the aircraft and a predicted airspeed 128 for the aircraft. The output parameter calculator 122 is implemented in any one or more of software, hardware, and firmware. The output parameter calculator calculates the predicted pressure 126 from the static pressure and a defined relationship between the static pressure and pressure altitude. In one example, the output parameter calculator calculates the predicted airspeed 128 using the predicted dynamic pressure inserted into the following equation:

This is expressed as dynamic pressure equal to half the air density multiplied by the square of the airspeed, where the air density is the known standard sea level air density corrected for the expected altitude.

Figure 5 shows an example of a neural network architecture used in the air data estimator 106. In the example of Figure 5, there is one neural network 520 shown in box 5, but in reality there are four such neural networks, each receiving the same input from the concatenation operation 504. The use of separate neural networks has been found to improve tuning ability and therefore accuracy.

In the example of FIG. 5, each neural network 520 has four layers: an input layer 506, two hidden layers 508, 510, and an output layer 512. The layers are fully connected, and each neural network 520 is a feedforward neural network. Note that in other examples, more than two hidden layers are used with increased regularization.

The input layer has at least x nodes, where x is the number of input values 500 from the avionics in the aircraft. In one example, the input values are engine data with thrust, static pressure, control surface position, aircraft mass, aircraft center of gravity, attitude, maneuver rate, and acceleration. In an example with 12 control surfaces, there are 20 input values and 20 nodes in the input layer 506. Those skilled in the art will understand that this example is not intended to be limiting and that other numbers of nodes may be used in the input layer depending on the number of input values used.

Input values are normalized by scaling them independently, in some cases using min-max normalization.

The normalized input values are concatenated into a single vector 504. Using concatenation into a single vector has been found to provide accurate results compared to the alternative approach of grouping the input values into multiple separate input vectors.

In some examples, hidden layers 508 and 510 have the same number of nodes as input layer 506 and are fully connected layers as described above. Note that it is not necessary for the number of nodes in hidden layers 508 and 510 to be the same as the number of nodes in the input layer. In some examples, when more than two hidden layers are used to arrive at a similar function, dropout layers are also used.

The output layer 512 has a single node for the parameter to be predicted. Each neural network 520 outputs a different parameter. In one example, there is one neural network to predict static pressure, one neural network to predict angle of attack 122, one neural network to predict sideslip 124, and one neural network to predict dynamic pressure 518.

The values at the nodes of the output layer 512 are denormalized 514 using the inverse of the normalization applied in operation 502. The results are predicted values of static pressure 516, dynamic pressure 518, angle of attack 122, and sideslip 124.

Each node in a neural network has an activation that defines how a weighted sum of the node's inputs is transformed into an output. Any suitable nonlinear activation function can be used, such as hyperbolic tangent (tanh) or rectified linear units for the input and hidden layers, and linear activation for the output layer.

In another example, the configuration of FIG. 5 is modified so that there is only one neural network and the output layer has one output node per output parameter. Those skilled in the art will understand that other neural network architectures may be used.

Figure 6a shows a flow diagram of an exemplary method for training a neural network for use in an air data estimator. Training data 322 may be available, such as having been calculated using the process of Figure 3. The training data may comprise tens of thousands or hundreds of thousands of training data pairs. Each training data pair comprises a set of ground truth values for air data parameters 120 and a corresponding set of values from the avionics in the aircraft (which may be simulated or empirical values).

The method of FIG. 6a is very computationally intensive and therefore may be performed in a distributed manner in the cloud. However, the use of distributed processing is not required.

A training example 600 is taken, which is a training data pair from the training data 322. A set of values from the avionics on board the aircraft is taken from the training data pair and, after normalization, is input to the neural network. The set of values is propagated forward through the neural network layers using forward propagation 602. Forward propagation comprises, at each node in a layer of the neural network, applying a function to the node's input value and applying the node's weight. The result of applying the function and weight, and optionally a bias, is an output value that is fed forward to the node in the next layer of the neural network, according to the connections between layers. The node weights are initially set to random values.

Forward propagation continues until the output layer of the neural network is reached. The output values of each node in the output layer are obtained and denormalized. The denormalized output values are compared to the corresponding ground truth values from the training data pairs. This comparison is performed by calculating a loss function 604, such as the sum of squared errors or other loss function. Examples of loss functions that may be used are:

In words, the loss is the sum over nodes n in a layer of the squared differences between the predicted value y output by a node given the inputs ^xn to the node and the node's weights ^w , and the ground truth value tn for the node.

Backpropagation is then computed 606, whereby the loss function is used to inform how the weights at the nodes in the neural network layers should be updated. Optimization is computed to determine the adjustments to make to the weights. This is done for each of the neural network layers during the backpropagation process.

At operation 608, a check is made to see if the process should be repeated using another training data pair. If so, the method returns to operation 600 to take another training example from the training data 322.

When the process is finished, the machine learning model is stored by storing the weights 610. Successively better models are continually saved to allow the best model to be recovered.

The decision as to whether to stop training in operation 608 is made according to one or more criteria, such as the number of training examples used, the time elapsed, the amount of weight change during recent backpropagation, or the amount of divergence in performance on the training and validation data sets.

Figure 6b shows a flow diagram of an exemplary method for validating a neural network for use in an air data estimator. Validation data 324 is available, which can be either data produced by the process of Figure 3 or flight test data, or a combination of these types of data. In the case of flight test data, air data parameter sensors are used on the aircraft's exterior to obtain pairs of ground truth air data parameter values and avionics values.

A validation example 612 is taken. The avionics values from the validation example are taken and, after normalization, input to the neural network input layer. Forward propagation 614 is performed as described with respect to FIG. 6A. Forward propagation results in values being output at the nodes in the output layer, which are compared to the ground truth values of the validation example. This comparison is performed by calculating a loss 616 using a loss function as described above. The result of the loss function is stored.

At decision point 618, a decision is made whether to repeat; if so, another validation example is taken and the process continues from operation 612. If the decision is not to repeat, an accuracy metric is calculated at operation 620. The decision to repeat at decision point 618 is made using criteria such as the number of validation examples processed.

To calculate the accuracy metric in operation 620, the standard deviation, mean, median, or mode of the calculated losses is calculated.

Once the neural network is trained and successfully validated, it is deployed to the aircraft. Successful validation occurs when the accuracy metric calculated in operation 620 exceeds a specified threshold.

In order to deploy a neural network on an aircraft, resource constraints must be met, including strict constraints on power, space, and time. The neural network should be low-power, i.e., capable of operating without drawing large amounts of power from the aircraft. The neural network should be compact, as space on an aircraft is at a premium. The neural network should be highly efficient to enable real-time operation, i.e., capable of predicting air data parameters in real time while the aircraft is flying complex maneuvers, where the air data parameters are expected to change very rapidly. In one example, real-time operation means calculating new values for the air data parameters approximately 80 times per second.

The inventors have developed a neural network architecture, as shown in Figure 5, that promotes low power, small space, and high-speed operation. Because the neural network of Figure 5 has only four layers, it is low power, small space, and fast. It is also possible to reduce the number of hidden layers 508, 510 from two to one. The inventors have found that such a configuration provides accurate results and improves compactness and power reduction.

The inventors have surprisingly found that the use of a recurrent neural network architecture is unnecessary. As a result, a feedforward neural network architecture with fully connected layers and no recursion is found to provide accurate results in a highly efficient manner. Because a recurrent neural network is not used, compactness is promoted. This reduces the size of the input space, as there is no need to examine temporal combinations.

In some embodiments, the neural network is deployed using hardware circuitry, such as a graphics processing unit or other parallel processing unit. In this case, the speed at which the neural network operates is increased compared to deploying the neural network using software.

FIG. 7 shows a flow diagram of an example method for using a deployable air data estimator 106 in an aircraft. The air data estimator 106 comprises a neural network trained using supervised learning. In one example, the neural network is trained as described with reference to FIG. 6a and validated as described with reference to FIG. 6b. The air data estimator 106 receives real-time inputs 700 from avionics onboard the aircraft. The real-time inputs comprise one or more of engine data such as thrust, static pressure, control plane position, mass, center of gravity, attitude, maneuver rate, and acceleration.

These inputs are normalized 702 by computing a min-max normalization or other normalization as described above with reference to FIG. 5. The normalized inputs are input to the input layer of the neural network in the air data estimator, and the forward propagation 704 is computed as described with reference to FIG. 6a, except that the weights are trained weights rather than random values.

The result of forward propagation 704 is an output value at a node in the output layer 512 of the neural network. The output value at the node in the output layer 512 is denormalized 706 by reversing the normalization operation from operation 702.

The unnormalized outputs, which are predictions of static and dynamic pressure, are input to the output parameter calculation 708, which calculates altitude and airspeed, as described above with reference to FIG. 4.

The air data estimator outputs parameter values 710, which are displayed 712, such as on a display in the cockpit. The parameter values 710 are also provided to control and/or management avionics 714 within the aircraft to automatically stabilize the aircraft or for other automated tasks.

An air data estimator and output parameter calculator were created and tested. Example test data are shown in Figures 8a, 8b, and 8c. The data shown in Figures 8a-8c was obtained from an air data estimator having a neural network architecture as illustrated in Figure 5, trained using training data calculated as described with reference to Figure 3. The neural network in the air data estimator was trained and validated as described with reference to Figures 6a-6b, and operated as described with reference to Figure 7. The plotted data in Figures 8a-8c compares simulated "true" maneuver predictions with the output from the air data estimator.

Figure 8a is a graph of sideslip angle estimated by the air data estimator versus the simulated ground truth sideslip angle. The x-axis of the graph represents time in seconds. The y-axis of the graph represents the aircraft's maneuver parameters. It can be seen that the sideslip angle predicted by the air data estimator matches the ground truth sideslip angle over most of the graph.

Figure 8b is a graph of the knot equivalent airspeed estimated by the air data estimator and the simulated ground truth knot equivalent airspeed. It can be seen that the knot equivalent airspeed predicted by the air data estimator generally follows the ground truth knot equivalent airspeed.

Figure 8c is a graph of the angle of attack estimated by the air data estimator versus the simulated ground truth angle of attack. It can be seen that the angle of attack predicted by the air data estimator closely matches the ground truth angle of attack throughout the graph.

The computer-implemented methods described herein may be performed in any software/firmware/hardware and using any suitable type of computing device, such as a microprocessor or other processor, and optionally, a graphics processing unit or other parallel processing device.

The computer-implemented methods described herein may be performed in any software/firmware/hardware and using any suitable type of computing device, such as a microprocessor or other processor, and optionally using a graphics processing unit or other parallel processing device.
The following is a summary of the claims as originally filed:
[C1]
An aircraft (100),
an air data estimator (106) comprising a machine learning model trained to predict air data parameter values from input values;
The air data estimator (106) is operable in real time on input values, where the input values are sensor data from sensors internal to the aircraft (100) only or are data derived from such sensor data.
[C2]
The aircraft (100) of C1 further comprising a flight control system (114), wherein the input values comprise values from the flight control system (114).
[C3]
The aircraft (100) of C2, wherein the input value comprises a static pressure value.
[C4]
The aircraft (100) of C2 or C3, wherein the input from the flight control system (114) comprises a control surface position.
[C5]
The aircraft (100) of any one of C1 to C4, comprising a digital engine control unit (108), wherein the input values comprise engine data from the digital engine control unit (108).
[C6]
6. The aircraft (100) of any one of C1 to C5, comprising a vehicle management system (110), wherein the input value comprises a mass of the aircraft (100) determined by the vehicle management system (110).
[C7]
The aircraft (100) of C6, wherein the input value comprises a center of gravity of the aircraft (100) determined by the fleet management system (110).
[C8]
The aircraft (100) of any one of C1 to C7, comprising a navigation subsystem and an inertial measurement unit that provide attitude data of the aircraft (100) within the input values.
[C9]
The aircraft (100) of C8, wherein the navigation subsystem and inertial measurement unit provide a maneuver rate within the input values.
[C10]
The aircraft (100) of any one of C8 to C9, wherein the navigation subsystem and inertial measurement unit provide an acceleration of the aircraft (100) within the input values.
[C11]
The aircraft (100) of any one of C1-10, wherein the air data parameter values comprise any one or more of angle of attack, sideslip, dynamic pressure, and static pressure.
[C12]
10. The aircraft of claim 9, wherein the air data parameter values comprise static pressure and dynamic pressure, and the aircraft comprises an output parameter calculator configured to calculate an altitude of the aircraft and an airspeed of the aircraft from the static pressure and dynamic pressure parameter values predicted by the air data estimator.
[C13]
The aircraft (100) of any one of C1 to C12, wherein the machine learning model is a neural network comprising an input layer, an output layer, and up to two hidden layers.
[C14]
The aircraft (100) of C11, wherein the neural network is a feed-forward neural network without recursion.
[C15]
The aircraft (100) of C11 or C12, wherein the neural network is implemented using hardware circuitry comprising a graphics processing unit or a parallel processing device.
[C16]
10. The aircraft of claim 1, wherein the machine learning model has been trained using labeled training data computed by randomly selecting values from a flight envelope of the aircraft and using the selected values to simulate the input values.
[C17]
A method performed in an aircraft (100), the method comprising:
1. A method comprising: using an air data estimator (106) comprising a machine learning model for predicting air data parameter values from input values in real time, wherein the input values are sensor data from sensors internal to the aircraft only or are data derived from such sensor data.
[C18]
The method of C17, wherein the input values comprise engine data, control surface position, static pressure, mass, center of gravity, attitude, steering rate, and acceleration.
[C19]
19. The method of claim 17 or 18, comprising concatenating the input values into a vector and inputting the vector into the machine learning model.
[C20]
20. The method of any one of C17 to 19, comprising normalizing the vectors before inputting them into the machine learning model.
[C21]
The method of any one of C17 to C20, comprising denormalizing output values output from the machine learning model.
[C22]
1. A computer-implemented air data estimator (106) comprising a machine learning model for predicting air data parameter values from input values in real time, wherein the input values are from avionics within the aircraft (100) such that the aircraft (100) does not require air data sensors mounted on the exterior of the aircraft (100).
[C23]
3. The computer-implemented air data estimator of claim 22, implemented using hardware circuitry, wherein the machine learning model is a neural network having less than three hidden layers.
[C24]
1. A computer-implemented method for training an air data estimator to predict values of air data parameters of an aircraft (100), the method comprising:
1. A computer-implemented method comprising: using supervised training to train a neural network in the air data estimator, the supervised training using training data pairs, each training data pair comprising a ground truth value of the air data parameter selected from a flight envelope of the aircraft and a corresponding simulated value from avionics in the aircraft, the simulated values being calculated using stored empirical data for an engine of the aircraft and empirical data for the aircraft obtained from wind tunnel testing, and the rules of computational fluid dynamics.
[C25]
The computer-implemented method of C24, wherein the neural network comprises multiple neural networks, one for each parameter to be predicted.

Claims (25)

An aircraft (100),
an air data estimator (106) comprising a machine learning model trained to predict air data parameter values from input values;
The air data estimator (106) is operable in real time on input values, where the input values are sensor data from sensors internal to the aircraft (100) only or are data derived from such sensor data. The aircraft (100) of claim 1, further comprising a flight control system (114), wherein the input value comprises a value from the flight control system (114). The aircraft (100) of claim 2, wherein the input value comprises a static pressure value. The aircraft (100) of claim 2 or claim 3, wherein the input value from the flight control system (114) comprises a control surface position. An aircraft (100) according to any one of claims 1 to 4, comprising a digital engine control unit (108), wherein the input values comprise engine data from the digital engine control unit (108). An aircraft (100) according to any one of claims 1 to 5, comprising a vehicle management system (110), wherein the input value comprises a mass of the aircraft (100) determined by the vehicle management system (110). The aircraft (100) of claim 6, wherein the input value comprises a center of gravity of the aircraft (100) determined by the transportation management system (110). An aircraft (100) as described in any one of claims 1 to 7, comprising a navigation subsystem and an inertial measurement unit that provide attitude data of the aircraft (100) within the input values. The aircraft (100) of claim 8, wherein the navigation subsystem and inertial measurement unit provide a maneuver rate within the input values. The aircraft (100) of claim 8 or claim 9, wherein the navigation subsystem and inertial measurement unit provide the acceleration of the aircraft (100) within the input values. An aircraft (100) according to any one of claims 1 to 10, wherein the air data parameter values comprise any one or more of angle of attack, sideslip, dynamic pressure, and static pressure. The aircraft (100) of claim 9, wherein the air data parameter values comprise static pressure and dynamic pressure, and the aircraft comprises an output parameter calculator configured to calculate an altitude of the aircraft (100) and an airspeed of the aircraft (100) from the static pressure and dynamic pressure parameter values predicted by the air data estimator (106). An aircraft (100) according to any one of claims 1 to 12, wherein the machine learning model is a neural network having an input layer, an output layer, and up to two hidden layers. The aircraft (100) of claim 11, wherein the neural network is a feedforward neural network without recursion. The aircraft (100) of claim 11 or 12, wherein the neural network is implemented using hardware circuitry comprising a graphics processing unit or a parallel processing device. The aircraft (100) of claim 1, wherein the machine learning model is trained using labeled training data calculated by randomly selecting values from the aircraft's flight envelope and using the selected values to simulate the input values. A method performed in an aircraft (100), the method comprising:
1. A method comprising: using an air data estimator (106) comprising a machine learning model for predicting air data parameter values from input values in real time, wherein the input values are sensor data from sensors internal to the aircraft only or are data derived from such sensor data. The method of claim 17, wherein the input values comprise engine data, control surface position, static pressure, mass, center of gravity, attitude, maneuver rate, and acceleration. The method of claim 17 or 18, comprising concatenating the input values into a vector and inputting the vector into the machine learning model. The method of any one of claims 17 to 19, further comprising normalizing the vector before inputting the vector into the machine learning model. The method of any one of claims 17 to 20, further comprising denormalizing the output values output from the machine learning model. A computer-implemented air data estimator (106) comprising a machine learning model for predicting air data parameter values from input values in real time, wherein the input values are from avionics within the aircraft (100) such that the aircraft (100) does not require air data sensors mounted on the aircraft's exterior. The computer-implemented air data estimator of claim 22, implemented using hardware circuitry, wherein the machine learning model is a neural network having fewer than three hidden layers. 1. A computer-implemented method for training an air data estimator to predict values of air data parameters of an aircraft (100), the method comprising:
1. A computer-implemented method comprising: using supervised training to train a neural network in the air data estimator, the supervised training using training data pairs, each training data pair comprising a ground truth value of the air data parameter selected from a flight envelope of the aircraft and a corresponding simulated value from avionics in the aircraft, the simulated values being calculated using stored empirical data for an engine of the aircraft and empirical data for the aircraft obtained from wind tunnel testing, and the rules of computational fluid dynamics. The computer-implemented method of claim 24, wherein the neural network comprises multiple neural networks, one for each parameter to be predicted.