Content area
Full Text
Abstract. This paper describes a case-study where we built and exercised a cloud computing framework with machine learning (ML) algorithms to improve the accuracy of Auxiliary Power Units (APU) health monitoring. An APU is a small turbo machine that flies on all commercial transport airplanes. The paper describes the objective of our study, sources of available data, the ETL scripts to populate the underlying HBase tables and two examples. In one example machine learning algorithms operating on multiple data sources produce useful insights to increase our ability to predict APU wear from 39% to 56%. In the second example, it increased our ability to predict shutdown events from 19% to 60%. This case-study illustrates the effectiveness of big data analytics and tools to discover additional insights that can further reduce operational interrupts arising from airborne equipment problems.
Keywords: Big Data Analytics, Machine Learning, System Health, Case Studies, Cloud Computing.
1 Introduction
In recent years, the aviation industry lias witnessed a steady increase in more data being collected by Aircraft Condition Monitoring Systems (ACMS). Data volumes ranging from 5 ~ 10 megabytes per flight hour (each aircraft) are routinely collected by onboard recorders and sent directly over airport Wi-Fi and GSM wireless networks without incurring the costs associated with ACARS messaging. Advances in IT and software that allow secure movement of data from airplanes provide an ideal framework for embedding statistical machine learning algorithms that can discover sweet-spots in global operations can feedback into day-to-day actions. This cloud computing based information network created by these connected aircrafts (a part of Industrial Internet) hold the potential of providing valuable knowledge needed to maintain profitability in an economically challenged civil aviation industry.
Technically, one of our study areas that can benefit most from current big data analytics is to reduce maintenance cost of high-value aerospace assets. For example, a recent GE article estimates a $250M savings in engine maintenance cost [1] is possible using insights gained from machine learning (ML), data mining and knowledge discovery. While the actual savings depend on specific aftermarket business policies, such case studies have been widely reported, clearly indicating the potential of data mining methods for discovering useful business insights from big data. In this paper, we describe our approach to...