Physics-informed Intelligent Analytics (PIA)

We propose the development of a toolset for Physics-informed Intelligent Automated Analytics (PIA) of missile simulations. PIA will become a universal tool kit to help accelerate all types of modeling and simulation (M&S) using data-driven surrogate models. When possible, PIA will leverage the physics underlying simulations to reduce the need for training data and improve surrogate model accuracy. PIA will automate analytics of large simulation outputs through Machine Learning (ML) based summarization and reporting. PIA will leverage the DoD authoritative source of truth and intelligently recommend most suitable model for their analytical objective. Use of surrogate models will help the DoD harmonize disparate models across programs. These models will improve M&S throughput by accelerating model chaining and enabling faster-than-real-time execution. PIA will utilize industry standards such as SysML and leverage cutting-edge technologies such as ML. PIA will automate fitting of Deep Neural Network (DNN) data-driven models to a wide variety of simulations. Other ML methods such as Meta Analysis will be used to automate simulation output analytics. PIA will implement advanced cloud-computing frameworks to streamline data-driven model fitting and execution. An efficient Model Scheduler and a robust idempotent Streaming Communicator will be integrated for fault-tolerant execution across variable networks. All innovations will be integrated into a cloud-capable and DoD security compliant toolset. We have built a team of leading experts in Digital Engineering (DE), ML and Systems Engineering to successfully develop PIA. Approved for Public Release | 21-MDA-11013 (19 Nov 21)