applications of machine learning in aviation · copyright © 2019 boeing. all rights reserved....

Copyright © 2019 Boeing. All rights reserved.

Engineering, Test & Technology

Applications of Machine Learning in Aviation

Presented at the ITA-MIT Workshop on Big Data Analytics for Air Transportation São José dos Campos, Brazil, August 20, 2019 Presenter: Ítalo Romani de Oliveira Authors: Chip Meserole, Sherry Yang Airspace Operational Efficiency Boeing Research & Technology


Boeing Research &Technology | Airline Operational Efficiency

Preceding work with Big Data in BR&T Brazil

2

Paper at the 20th International Conference of Intelligent Transportation Systems (ITSC) 2017, In partnership with University of Brasilia



Outline

n  Introduction n Motivation n Challenges

n  Background

n  Research areas n Robustness n Deep sense learning n Recent examples

n Summary



Machine Learning (ML) is defined by widely accepted definitions as:

A computer program is said to learn from experience with respect to some class of tasks and performance measure if its performance at those tasks improves with experience.

— Tom Mitchell or:

Field of study that gives computers the ability to learn without being explicitly programmed.

— Arthur Samuel, ca. 1959



Supervised & Unsupervised Learning Both Provide–

System Performance Enhancements

Smarter Devices

Fewer Errors Better Efficiency Increasing Accuracy

Yielding



Challenges for Machine Learning in Aviation

n  Non-stationary stochastic processes

n  Anomaly detection

n  Learning in on-line settings

n  Others include:

n  Real-time operation

n  Competing systems



A Sample of Machine Learning Applications Profile 50 Years of History

Time

Low

Mathematics Problems

1980s 1990s 2000s 2010s 2020s

Degrees of Improvement

Low

Societal Planning

Human-like performance

Medium

Low

Com

plex

ity Difficult

High

Stock Trends

Very High Advanced Robotics

Very High

Surgical Assistant

Autonomous Cars

Very High



Importantly, ML Cannot Ignore Training at the Corners but … Methods Do Not Extrapolate

Frequency of Occurrence

High

Con

sequ

ence

of

Occ

urre

nce

Low

Low High



High-Stakes Applications Require Robust AI

In aviation, best practice for AI / ML robustness must guarantee safety, even with the presence of:

n  Incomplete models n  Unmodeled phenomena



Why is Unmodeled Phenomena a hard fundamental problem?

1)  It is impossible to model everything n  Qualification Problem:

n  It is impossible to enumerate all of the preconditions for an action

n  Ramification Problem: n  It is impossible to enumerate all of the

implicit consequences of an action

n  Fundamental theorem of machine learning

error rate∝model complexity/sample size  n  Corollary:

n  If sample size is small, the model should be simple n  We must deliberately oversimplify our models !

2)  It is important NOT to model everything



Deep Sense Learning (DSL)

Objective

n  Need for robust scene understanding in cluttered environments and data analytics with few labeled data

n  Our objective is to dramatically reduce false alarms for autonomous recognition and learn with 1000x less labeled data

Benefits

Autonomous navigation

Target recognition

Novelty detection

Performance

VGG-19

deep net Huge improvement with DSL

n  For 50 labeled samples per class, DSL reduced errors from 70% for deep learning to only 18%



Deep Sense Learning technology explains the inner workings of complex machine learning models

Explainable Machine Learning

DSL explains machine learning mistakes and how to correct them

Challenge High-performance ML models are black boxes

Innovation Explainable Traces (X-trace) provide fine-grained view of ML decisions

Decompose input into hierarchy of decision-relevant attributes Approach


Aircraft Trajectory Prediction Using Advanced Analytics

Night OWL TP

Server

Radar Data

Flight Plan Data

Weather Data

One way data

HMI Server

Area Controller

Tablet

Approach Controller

Tablet

User Changes

User Changes One way data

One way data

Runway Threshold

Merge Point

Night OWL TP



Some recent references

In: Proceedings of 2018 IEEE/AIAA 37th Digital Avionics Systems Conference (DASC) (Available in IEEExplore)

Saccone, G.; Mead, R.: Viability of the Use of Aircraft-generated Intent Data for Air Traffic Management

J. Lopez Leones et al.: Advanced Flight Efficiency Key Performance Indicators to support Air Traffic Analytics: Assessment of European flight efficiency using ADS-B data

A. Muñoz Hernandes et al.: Data-driven Aircraft Trajectory Predictions using Ensemble Meta-Estimators

I. Garcia-Miranda et al.: AIRPORTS Metrics: A Big Data application for computing flights' performance indexes based on flown trajectories



Summary

n  The AI/ML science in is rapidly evolving

n  The scientific evolution goes in parallel to the evolution of autonomous transportation systems

n  Robustness is essential, or ML will produce unexpected outcomes

applications of machine learning in aviation · copyright © 2019 boeing. all rights reserved....

Documents