Methods and systems for inferred information propagation for aircraft prognostics

A prognosis system for predicting a failure of a component of an aircraft in real-time, said prognosis system comprising:

• a memory for storing data;
• and
• a processor in communication with said memory, said processor programmed to:
• receive respective original time-series of data points for the component as an input from a plurality of sensors on the aircraft, wherein the respective original time-series of data points are derived from data generated in-flight, and wherein the respective original time-series of data points comprise variables including at least one current feature, at least one voltage feature, and at least one accelerometer feature;
• preprocess the respective original time-series to define a plurality of overlapping time windows over which the respective original time-series are divided, thereby defining respective subsets of data for each of the plurality of overlapping time windows, wherein the respective subsets of data for each of the plurality of overlapping time windows comprises two or more data points;
• compute a Mutual Information (MI) value for each pair of the variables within each of the respective subsets of data of the respective original time-series;
• construct a time-series of relationship graphs having weights corresponding to the MI values and ordered over the plurality of overlapping time windows, wherein nodes of the relationship graphs correspond to the variables and edges of the relationship graphs correspond to the weights corresponding to the MI values;
• cluster the relationship graphs for each of the plurality of overlapping time windows, thereby defining a time-series of clusters, wherein the relationship graphs are clustered based on statistical dependencies of the variables of the relationship graphs;
• detect a change over time in at least one of the weights or a structure of the time-series of clusters, wherein the change is detected in-flight, and wherein the change corresponds to increased coupling between at least two of the variables, indicating a change in a condition of the component of the aircraft;
• and
• identify, in-flight to a user, the change as a feature corresponding to at least two of the respective original time-series from which the change originated and alert the user to a potential predicted failure of the component of the aircraft based on the feature.

One or more non-transitory computer-readable storage media having computer-readable instructions encoded thereon, wherein when executed by a processor, said computer-readable instructions cause the processor to:

• receive respective original time-series of data points for a component of an aircraft as an input from a plurality of sensors on the aircraft, wherein the respective original time-series of data points are derived from data generated in-flight, and wherein the respective original time-series of data points comprise variables including at least one current feature, at least one voltage feature, and at least one accelerometer feature;
• preprocess the respective original time-series to define a plurality of overlapping time windows over which the respective original time-series are divided, thereby defining respective subsets of data for each of the plurality of overlapping time windows, wherein the respective subsets of data for each of the plurality of overlapping time windows comprises two or more data points;
• compute a Mutual Information (MI) value for each pair of the variables within each of the respective subsets of data of the respective original time-series;
• construct a time-series of relationship graphs having weights corresponding to the MI values and ordered over the plurality of overlapping time windows, wherein nodes of the relationship graphs correspond to the variables and edges of the relationship graphs correspond to the weights corresponding to the MI values;
• cluster the relationship graphs for each of the plurality of overlapping time windows, thereby defining a time-series of clusters, wherein the relationship graphs are clustered based on statistical dependencies of the variables of the relationship graphs;
• detect a change over time in at least one of the weights or a structure of the time-series of clusters, wherein the change is detected in-flight, and wherein the change corresponds to increased coupling between at least two of the variables, indicating a change in a condition of the component of the aircraft;
• and
• identify, in-flight to a user, the change as a feature corresponding to at least two of the respective original time-series from which the change originated and alert the user to a potential predicted failure of the component of the aircraft based on the feature.