A Plug-and-Play (PnP) Tool based on Online Machine Learning for Real-Time Monitoring and Control of Mechanical Systems

The proposed effort aims to develop and demonstrate a plug-and-play (PnP) tool/platform based on online neural network (NN) learning and modeling for real-time monitoring, prognostics, and control of mechanical systems. The salient aspects of the proposed solution are: (1) “around-the-clock” learning and identification of mechanical system enabling continuous dynamics tracking and model updating; (2) feature selection to extract the most representative inputs/features for compact model structure; (3) model predictive control (MPC) for real-time optimization and control synthesis; and (4) innovation in software-hardware architecture to streamline NN learning, model construction, MPC on a hybrid and embedded platform to meet demanding Size, Weight, and Power (SWaP) requirements in HUMS. In Phase I, key components including a NN-based online ML module, an input/feature selection module, a MPC module, and an embedded system architecture will be developed. Feasibility will be demonstrated via case studies of US Army interest, including real-time situational response demonstrations in a controlled simulated environment using datasets from both synthetic and real-world operations. The Phase II effort will focus on capability extension, algorithm optimization, software integration, extensive technology validation and demonstration, and technology insertion into Army’s workflow.