The Boeing Company

James M Ethington

Liessman E Sturlaugson

Jason M Keller

Nile Hanov

Timothy Joseph Wilmering

Mark H Boyd

James Schimert

Lee Wang

A method is provided that includes accessing a multivariate time series of flight data for an aircraft, and iteratively performing runs of genetic programming on groups of the sensors. A population of computer programs is randomly generated from a selected group of the plurality of sensors, and primitive functions selected from a library of primitive functions. The population is iteratively transformed into new generations of the population, and includes sub-rankings of the group of sensors based on a quantitative fitness determined according to selected fitness criterion. A ranking of the group of sensors from the sub-rankings of the group of sensors is produced. An aggregate ranking of the plurality of sensors is produced from the ranking of the group of sensors over a plurality of iterations. And the subset of sensors is selected from the aggregate ranking of the plurality of sensors, and according to selected optimization criterion.

Automated feature generation for sensor subset selection

A method is provided for diagnosing a failure on an aircraft that includes aircraft systems and monitors configured to report effects of failure modes of the aircraft systems. The method includes receiving a fault report that indicates one or more of the monitors that reported the effects of a failure mode in an aircraft system of the aircraft systems, and accessing a fault pattern library that describes relationships between possible failure modes and patterns of those of the monitors configured to report the effects of the possible failure modes. The method also includes diagnosing the failure mode of the aircraft system from the one or more of the monitors that reported, and using the fault pattern library and a greedy selection algorithm, determining a maintenance action for the failure mode; and generating a maintenance message including at least the maintenance action.

System and method for contextually-informed fault diagnostics using structural-temporal analysis of fault propagation graphs

A data processing system may include instructions stored in a memory and executed by a processor to categorize a plurality of systems into clusters using an unsupervised machine learning method to analyze repair data parameters of a historical dataset relating to the plurality of systems. The system may assign a repair forecast to each cluster, and may generate a system repair forecasting model using selected predictor variables, the historical data set, and the repair forecasts according to a supervised machine learning method. The selected predictor variables may correspond to a mathematical combination of operational data parameters in the historical dataset.

Machine learning based repair forecasting

In an example, a method for determining whether to perform aircraft maintenance is described. The method comprises selecting groupings of sensors and/or parameters associated with an aircraft type. The method comprises receiving feature data that corresponds to each grouping of sensors and/or parameters. The method comprises determining, from the feature data, values for predetermined operational metrics. The method comprises comparing the values to values for predetermined operational metrics that correspond to at least one other flight of the aircraft. The method comprises determining, based on comparing the values for the predetermined operational metrics, values of additional operational metrics. The method comprises training a machine learning model using at least the values for the additional operational metrics. The method comprises providing, based on an output of the machine learning model, an instruction to perform maintenance on one or more of the plurality of portions of the aircraft or another aircraft.

Methods and systems for performing aircraft maintenance

A vehicle maintenance scheduling and fault monitoring apparatus includes a vehicle system maintenance rules generation module and a vehicle system fault detection module. The rules generation module determines a correlation between pairs of precedent historical vehicle fault data, of historical time-stamped vehicle fault data, and a subsequent different historical vehicle fault data, and generates vehicle system maintenance rules based on the correlation determined for the pairs of the precedent historical vehicle fault data and the subsequent different historical vehicle fault data. The fault detection module monitors faults of the vehicle system, determines an imminent occurrence of a subsequent vehicle fault, based on application of the vehicle system maintenance rules to the plurality of time-stamped precedent vehicle fault data, and generates a maintenance report corresponding to the imminent occurrence of the subsequent vehicle fault so that proactive maintenance is performed on the vehicle system.

Apparatus and method for vehicle maintenance scheduling and fault monitoring

Method and apparatus for detecting anomalous flights. Embodiments collect sensor data from a plurality of sensor devices onboard an aircraft during a flight. Feature definitions are determined, specifying a sensor device and an algorithm for deriving data values from sensor data collected from the device. Embodiments determine whether anomalous activity occurred during the flight using an anomaly detection model. An anomaly is detected including at least one of (i) a contextual anomaly where a data instance of a plurality of data instances is anomalous relative to a specific context, or (ii) a collective anomaly where two or more data instances are anomalous relative to a remainder of the plurality of data instances, even though each of the two or more data instances is not anomalous in and of itself. A report specifying a measure of the anomalous activity for the flight is generated.

On-board networked anomaly detection (ONAD) modules

Methods include qualifying a qualification dataset of fatigue-related parameters calculated from data collected during a first group of flights of a first aircraft that experienced a replacement of a structural component during heavy maintenance. Methods include receiving a qualified selection of the fatigue-related parameters and verifying a verification dataset of the qualified selection of the fatigue-related parameters calculated from data collected during a second group of flights of a second aircraft that experienced heavy maintenance without replacement of the structural component. Methods include receiving a set of verified and qualified fatigue-related parameters and building a predictive model for structural repair during heavy maintenance with a training dataset of the verified and qualified fatigue-related parameters calculated from data collected during additional flights of the fleet. Hybrid feature selection systems also are disclosed.

Prediction methods and systems for structural repair during heavy maintenance of aircraft

Hybrid feature selection methods include methods of creating a predictive model for valve performance in a fleet of aircraft. Methods include qualifying a qualification dataset of valve-related parameters calculated from data collected during a first series of flights at least before and after a non-performance event of a valve. Methods include receiving a qualified selection of the valve-related parameters and verifying a verification dataset of the qualified selection of the valve-related parameters calculated from data collected during a second series of flights. Methods include receiving a set of verified and qualified valve-related parameters and building a predictive model for valve non-performance with a training dataset of the verified and qualified valve-related parameters calculated from data collected during additional flights of the fleet.

Hybrid feature selection for performance prediction of fluid control valves

Method and apparatus for detecting anomalous flights. Embodiments collect sensor data from a plurality of sensor devices onboard an aircraft during a flight. A plurality of feature definitions are determined, where a first one of the feature definitions specifies one or more of the plurality of sensor devices and an algorithm for deriving data values from sensor data collected from the one or more sensor devices. Embodiments determine whether anomalous activity occurred during the flight using an anomaly detection model, where the anomaly detection model describes a pattern of normal feature values for at least the feature definition, and comprising comparing feature values calculated from the collected sensor data with the pattern of normal feature values for the first feature definition. A report specifying a measure of the anomalous activity for the flight is generated.

A vehicle health monitoring system including a rules generation module that receives historical time-stamped vehicle fault data for a vehicle system, determines a correlation between pairs of a precedent historical vehicle fault data, of the historical time-stamped vehicle fault data, and a subsequent different historical vehicle fault data, of the historical time-stamped vehicle fault data, and generates maintenance rules based on the correlation determined for the pairs. A fault detection module of the system monitors faults of the vehicle system, where the vehicle faults include a plurality of time-stamped precedent vehicle fault data, applies the rules to the plurality of time-stamped precedent vehicle fault data, determines an imminent occurrence of a subsequent vehicle fault, and generates a maintenance report corresponding to the imminent occurrence of the subsequent vehicle fault so that proactive maintenance is performed on the vehicle system.

Systems and methods of the present disclosure include determining a performance status of a selected component in an aircraft. An ensemble of related machine learning models is applied to feature data extracted from flight data of the aircraft. Each model produces a positive score and a complementary negative score related to performance of the selected component. The positive scores are weighted based on the false positive rates of the models and the negative scores are weighted based on the false negative rates of the models. The weighted positive scores are combined, e.g., by averaging, and the weighted negative scores are combined, e.g., by averaging. The performance status of the selected component is determined as one of a positive category, a negative category, or an unclassified category based on the values of the combined weighted positive scores and the combined weighted negative scores.

Advanced analytic methods and systems utilizing trust-weighted machine learning models

Hybrid feature selection methods include methods of creating a predictive model for structural repair during heavy maintenance in a fleet of aircraft. Methods include qualifying a qualification dataset of fatigue-related parameters calculated from data collected during a first group of flights of a first aircraft that experience a replacement of a structural component during heavy maintenance. Methods include receiving a qualified selection of the fatigue-related parameters and verifying a verification dataset of the qualified selection of the fatigue-related parameters calculated from data collected during a second group of flights of a second aircraft that experienced heavy maintenance without replacement of the structural component. Methods include receiving a set of verified and qualified fatigue-related parameters and building a predictive model for structural repair during heavy maintenance with a training dataset of the verified and qualified fatigue-related parameters calculated from data collected during additional flights of the fleet.

Predictive aircraft maintenance systems and methods are disclosed. Predictive maintenance methods may include extracting feature data from flight data collected during a flight of the aircraft, applying an ensemble of related classifiers to produce a classifier indicator for each classifier of the ensemble of classifiers, aggregating the classifier indicators to produce an aggregate indicator indicating an aggregate category of a selected component for a threshold number of future flights, and determining the performance status of the selected component based on the aggregate indicator. The classifiers are each configured to indicate a category of the selected component within a given number of flights. The given number of flights for each classifier is different. The threshold number of future flights is greater than or equal to the maximum of the given numbers of the classifiers. Predictive maintenance systems may include modules configured to extract feature data, classify feature data, and aggregate classifications.

Growing community of inventors

St. Charles, MO, United States of America

James M Ethington