Using Big Data and predictive machine learning in aerospace test environments

It is estimated that in 2012 most mid-size companies in the USA generate the equivalent data of the US Library of Congress in 1 year. As a company, Wal-Mart creates the equivalent of 50 million filing cabinets worth of data every hour. While these numbers seem incredible, the trend for most companies is an increasing volume of data generation and storage. Test Data generated by Automatic Test Equipment (ATE) in R&D, manufacturing and Repair environments is no exception to this increased volume of data. The challenge of this enormous amount of Test Data is how to provide people with effective ways to make decisions from it. Data visualization through charts, graphs and reports has been, historically, one of the more effective ways to provide actionable intelligence because humans can readily make decisions based on patterns and comparisons. But as data volume goes up, even this method is reaching its limits. When one starts to combine large datasets like Manufacturing Test Data and Repair Data together, data visualization becomes problematic. More sophisticated algorithmic, machine learning and predictive approaches become critical. In this paper, we will explore the experiences of using predictive algorithms on "Big Data" from both Manufacturing Test and Repair Test environments in the complex mission critical aerospace industry. By effectively using datasets from different functional areas, we will be looking at applying SPC techniques to answer new questions about the correlation of Repair test data and manufacturing data with the end goal to predict number of returns in the future and minimize product escapes.