(3) Machine Learning for Predictive Maintenance

The US Navy owns and maintains the one of the largest fleets of military aircraft in the world. Maintaining and supplying these aircraft is both time and resource intensive. Virtualitics proposes to produce a proof of concept during Phase I that demonstrates the feasibility of using machine learning to predict maintenance needs for a specific aircraft component. Using machine learning techniques in natural language processing, Virtualitics proposes algorithmic mining of data sources, such as maintenance logs, and the use of several nonlinear classifiers, such as neural networks, gradient boosting classifiers, or random forests, and ensemble the models to create the final prediction.In conjunction, we also propose to automate the calculation of several diagnostics and metrics via python script. Much of the data processing and analysis done by hand by analyst could be automated to run not only thousands of times more quickly, but also with greater accuracy and fidelity. Automating manual and time-consuming workflows gives analysts more time to undertake analyses with higher risk and greater reward, including assessing the impact of predictive maintenance.