Multi-Target High Probability of Kill Weapons Engagement - MP 129-11 Navy SBIR FY2011.3 Sol No.: Navy SBIR FY2011.3 Topic No.: N113-176 Topic Title: Multi-Target High Probability of Kill Weapons Engagement - MP 129-11 Proposal No.: N113-176-0115 Firm: Metron, Inc. 1818 Library Street Suite 600 Reston, Virginia 20190 Contact: Lauren Klak Phone: (619) 727-4113 Web Site: www.metsci.com Abstract: The torpedo threat to U.S. and coalition naval forces is real and growing. Incorporating machine learning into the fire control piece for the Torpedo Warning System (TWS) can help to increase the probability of kill for the large number of possible torpedo threats. Metron, Inc. proposes a unique solution using a machine learning algorithm to provide better performance across a wider solution space than the current program-of-record approach. One of the primary machine learning approaches we are considering is the use of the approximate dynamic programming (DP) algorithm. The goal of our algorithm is to maximize the probability of kill in a setting with multiple concurrent hostile torpedoes. The stochastic disturbance in the algorithm will take the form of a Gaussian Process Model with squared exponential covariance, accounting for the dynamic and uncertain information surrounding TWS. For the Phase I Option, Metron will produce a design to integrate the machine learning algorithm into the Naval Simulation System (NSS) to demonstrate our technology. Finally, Metron will run several simulations and gather metrics such as probability of kill and probability of false alarm for Countermeasure Anti-Torpedoes (CATs). Benefits: The techniques for fire control developed in this SBIR could be applied to several projects in the scientific, commercial, and DoD communities. In addition to the integration into TWS, there are practical extensions of a machine learning approach using the approximate DP algorithm. One interesting application would be to use the algorithm to detect and predict potential earthquake sites by modeling volcanic and tectonic processes. Many attempts to model volcanic processes still use relatively simple models such as a point pressure source in a uniform half-space. Although the simple models are valuable, updated models that depend on the variation of elastic properties and geometry could be implemented using the DP algorithm. This has the advantage of providing the full a posteriori probability density for the model parameters. Our technology is also applicable to automation efforts in the commercial rail industry, robotics industry, and commercial power industry. Return