WATCHER: Machine Learning and Defect Analysis for Detection of Latent Errors in Combat System Software

The opportunity for programs like AEGIS is not only in their voluminous test data, but also the vast amounts of data from prior software quality assurance (SQA) activities. By applying machine learning, these data will allow AEGIS and other programs to: 1. Detect latent errors in combat systems based on extremely complex software, 2. Develop meaningful test coverage metrics for MIL-STD-882E risk analysis, 3. Proactively target automated robustness testing to address residual risks, 4. And, ultimately, to improve cost effectiveness of Navy integration efforts. Phase I of our project will define pattern-recognition algorithms inspired by machine-learning techniques from natural-language processing, deep learning, and defect analysis. We will determine the feasibility of these algorithms based on representative data sets provided by the AEGIS program office and available via open source repositories, identified in collaboration with AEGIS program personnel at Lockheed Martin Rotary and Mission Systems (LM RMS). These data sets may be augmented with test data already available to the Switchboard automated-testing platform maintained by Edge Case Research. In collaboration with LM RMS, we will produce a capabilities description document and a roadmap showing how AEGIS SQA processes can adopt our technology starting with prototypes in Phase II.