US-12623793-B2 - Machine-learning-based flight data parameter estimation

Abstract

An aircraft includes airstream sensors configured to generate airstream data representing measurements of a diverse set of airstream parameters. The aircraft also includes processor(s) configured to provide input data to multiple machine-learning (ML) regression models to generate output data values. The input data provided to each ML regression model is based on the airstream data and represents two or more different airstream parameters, and each output data value represents an estimated value of a flight data parameter. The processor(s) are configured to provide output data values from two or more of the ML regression models to an ensembler to generate a unified output value indicating a unified estimate of the flight data parameter. The processor(s) are configured to generate an output based on a comparison of a procedurally determined computed value of the flight data parameter and the unified output value.

Inventors

• Zachary Robert Meves
• David Felipe Leguizamo

Assignees

• THE BOEING COMPANY

Dates

Publication Date

20260512

Application Date

20240325

Claims (20)

1. 1 . An aircraft comprising: a plurality of airstream sensors configured to generate airstream data representing measurements of a diverse set of airstream parameters; and one or more processors configured to: provide input data to multiple machine-learning (ML) regression models to generate two or more output data values, wherein the input data provided to each ML regression model is based on the airstream data and represents two or more different airstream parameters, and wherein each output data value represents an estimated value of a flight data parameter; provide output data values from two or more of the ML regression models as input to an ensembler to generate a unified output value indicating a unified estimate of the flight data parameter; obtain a computed value of the flight data parameter, wherein the computed value is determined procedurally at a primary flight computer based on at least a subset of the airstream data; and generate an output based on a comparison of the computed value and the unified output value.
2. 2 . The aircraft of claim 1 , wherein the one or more processors are further configured to obtain configuration data indicating a configuration setting of the aircraft, and wherein the input data of at least one of the ML regression models is further based on the configuration data.
3. 3 . The aircraft of claim 2 , wherein the configuration data indicates a flap setting, a landing gear position, or both.
4. 4 . The aircraft of claim 1 , wherein the one or more processors are further configured to obtain inertial sensor data, and wherein the input data of at least one of the ML regression models is further based on the inertial sensor data.
5. 5 . The aircraft of claim 1 , wherein the one or more processors are further configured to provide input data to one or more ML classification models to generate a classification output indicating whether one or more values of the airstream data is considered to be valid.
6. 6 . The aircraft of claim 1 , wherein the ML regression models include: a first ML regression model configured to generate a first estimated value of the flight data parameter based on a first subset of the airstream parameters; and a second ML regression model configured to generate a second estimated value of the flight data parameter based on a second subset of the airstream parameters, wherein the first subset of the airstream parameters includes one or more airstream parameters absent from the second subset of the airstream parameters.
7. 7 . The aircraft of claim 1 , wherein the ML regression models include: a first ML regression model having a first model architecture; and a second ML regression model having a second model architecture different than the first model architecture.
8. 8 . The aircraft of claim 1 , wherein the plurality of airstream sensors include one or more angle of attack sensors, one or more total pressure sensors, one or more temperature sensors, and one or more static pressure sensors, and wherein the flight data parameter corresponds to aircraft angle of attack, Mach number, aircraft sideslip angle, or altitude.
9. 9 . The aircraft of claim 1 , wherein one or more of the ML regression models is trained using training data representing an entire flight envelope of the aircraft.
10. 10 . The aircraft of claim 1 , further comprising a flight deck display, wherein the output is used to generate an indicator for the flight deck display.
11. 11 . The aircraft of claim 1 , wherein the computed value is provided to a flight deck display.
12. 12 . A method comprising: obtaining airstream data representing measurements of a diverse set of airstream parameters from a plurality of airstream sensors of an aircraft; providing input data to multiple machine-learning (ML) regression models to generate two or more output data values, wherein the input data provided to each ML regression model is based on the airstream data and represents two or more different airstream parameters, and wherein each output data value represents an estimated value of a flight data parameter; providing output data values from two or more of the ML regression models as input to an ensembler to generate a unified output value indicating a unified estimate of the flight data parameter; obtaining a computed value of the flight data parameter, wherein the computed value is determined procedurally at a primary flight computer based on at least a subset of the airstream data; and generating an output based on a comparison of the computed value and the unified output value.
13. 13 . The method of claim 12 , further comprising obtaining configuration data indicating a configuration setting of the aircraft, and wherein the input data of at least one of the ML regression models is further based on the configuration data.
14. 14 . The method of claim 13 , wherein the configuration data indicates a flap setting, a landing gear position, or both.
15. 15 . The method of claim 12 , further comprising obtaining inertial sensor data, and wherein the input data of at least one of the ML regression models is further based on the inertial sensor data.
16. 16 . The method of claim 12 , further comprising providing input data to one or more ML classification models to generate a classification output indicating whether one or more values of the airstream data is considered to be valid.
17. 17 . The method of claim 12 , wherein the ML regression models include: a first ML regression model configured to generate a first estimated value of the flight data parameter based on a first subset of the airstream parameters; and a second ML regression model configured to generate a second estimated value of the flight data parameter based on a second subset of the airstream parameters, wherein the first subset of the airstream parameters includes one or more airstream parameters absent from the second subset of the airstream parameters.
18. 18 . The method of claim 12 , wherein the ML regression models include: a first ML regression model having a first model architecture; and a second ML regression model having a second model architecture different than the first model architecture.
19. 19 . The method of claim 12 , further comprising: providing additional input data to multiple additional ML regression models to generate two or more additional output data values, wherein the additional input data provided to each additional ML regression model is based on the airstream data and represents two or more different airstream parameters, and wherein each additional output data value represents an estimated value of an additional flight data parameter; providing additional output data values from two or more of the additional ML regression models as input to an additional ensembler to generate an additional unified output value indicating a unified estimate of the additional flight data parameter; and obtaining a computed value of the additional flight data parameter, wherein the computed value of the additional flight data parameter is determined procedurally based on at least a subset of the airstream data, wherein the output is further based on a comparison of the computed value of the additional flight data parameter and the additional unified output value.
20. 20 . A line-replaceable unit comprising: one or more processors configured to: obtain airstream data representing measurements of a diverse set of airstream parameters from a plurality of airstream sensors of an aircraft; provide input data to multiple machine-learning (ML) regression models to generate two or more output data values, wherein the input data provided to each ML regression model is based on the airstream data and represents two or more different airstream parameters, and wherein each output data value represents an estimated value of a flight data parameter; provide output data values from two or more of the ML regression models as input to an ensembler to generate a unified output value indicating a unified estimate of the flight data parameter; obtain a computed value of the flight data parameter, wherein the computed value is determined procedurally at a primary flight computer based on at least a subset of the airstream data; and generate an output based on a comparison of the computed value and the unified output value.

Description

FIELD OF THE DISCLOSURE The present disclosure is generally related to estimation of flight data parameters using machine-learning. BACKGROUND Conventional methods for determining calculated flight data parameters of aircraft assume proper calibration and operation of sensors. Additionally, many flight data parameters are calculated based on data from two or more sensors (e.g., temperature and static pressure). Errors can be introduced in such flight data parameters due to issues with any of these sensors. Redundancies are generally built in to limit the effect of improperly calibrated sensors; nevertheless, there is always a desire to further improve reliability of flight data systems for aircraft. Adding additional redundant sensors is one way that reliability can be further improved. However, additional redundant sensors add cost, weight, and complexity to the aircraft. SUMMARY According to one implementation of the present disclosure, an aircraft includes a plurality of airstream sensors configured to generate airstream data representing measurements of a diverse set of airstream parameters. The aircraft also includes one or more processors configured to provide input data to multiple machine-learning (ML) regression models to generate two or more output data values. The input data provided to each ML regression model is based on the airstream data and represents two or more different airstream parameters, and each output data value represents an estimated value of a flight data parameter. The one or more processors are further configured to provide output data values from two or more of the ML regression models as input to an ensembler to generate a unified output value indicating a unified estimate of the flight data parameter. The one or more processors are also configured to obtain a computed value of the flight data parameter, where the computed value is determined procedurally based on at least a subset of the airstream data. The one or more processors are further configured to generate an output based on a comparison of the computed value and the unified output value. According to another implementation of the present disclosure, a method includes obtaining airstream data representing measurements of a diverse set of airstream parameters from a plurality of airstream sensors of an aircraft. The method also includes providing input data to multiple ML regression models to generate two or more output data values. The input data provided to each ML regression model is based on the airstream data and represents two or more different airstream parameters, and each output data value represents an estimated value of a flight data parameter. The method further includes providing output data values from two or more of the ML regression models as input to an ensembler to generate a unified output value indicating a unified estimate of the flight data parameter. The method also includes obtaining a computed value of the flight data parameter, where the computed value is determined procedurally based on at least a subset of the airstream data. The method further includes generating an output based on a comparison of the computed value and the unified output value. According to another implementation of the present disclosure, a line-replaceable unit includes one or more processors configured to obtain airstream data representing measurements of a diverse set of airstream parameters from a plurality of airstream sensors of an aircraft. The one or more processors are further configured to provide input data to multiple ML regression models to generate two or more output data values. The input data provided to each ML regression model is based on the airstream data and represents two or more different airstream parameters, and each output data value represents an estimated value of a flight data parameter. The one or more processors are also configured to provide output data values from two or more of the ML regression models as input to an ensembler to generate a unified output value indicating a unified estimate of the flight data parameter. The one or more processors are further configured to obtain a computed value of the flight data parameter, where the computed value is determined procedurally based on at least a subset of the airstream data. The one or more processors are also configured to generate an output based on a comparison of the computed value and the unified output value. The features, functions, and advantages described herein can be achieved independently in various implementations or may be combined in yet other implementations, further details of which can be found with reference to the following description and drawings. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a diagram that illustrates an aircraft including a line-replaceable unit configured to use ML models to estimate flight data parameters according to a particular implementation. FIG. 2 is a diagram illustrating aspects of operati