Underwater Munitions Expert System to Predict Mobility and Burial

Dr. Sarah Rennie | Johns Hopkins

MR-2227

Objective

A comprehensive model for predicting the location and possible burial of underwater munitions is required to advise site managers as they plan monitoring or clean-up activities. In any real-world scenario, the exact types and initial locations of the munitions as well as the environmental conditions will not be exactly known. A meaningful approach for predicting munitions’ locations and burial extent will be probabilistic in nature. The objective of this project was to build a demonstration computer tool implementing a probabilistic system that would predict patterns of migration, exposure and aggregation for underwater munitions.

Back to Top

Technical Approach

In order to have developed this expert system, a number of hydrodynamic and geological processes must be understood and related to the interaction of munitions with the underlying sediments and the environmental forces. Simple models relating causal forces acting on the underwater munitions and the associated sediment responses have been developed to predict scour burial and motion initiation. Recent work has extended the models to account for additional factors that are particularly important under wave-driven conditions. These models are used in the construction of a probabilistic Bayesian network forming the core of (Underwater Munitions Expert System) UnMES.

Back to Top

Results

Implemented improvements to UnMES include models for: the role of munitions density on burial and migration; the effects of oscillatory flow; the effects of the angle of bottom flow; acceleration effects on initiation of motion by the incorporation of an inertial factor for mobility threshold; and preliminary modeling of migration distance. In addition, model predictions are compared to field data obtained during Strategic Environmental Research and Development Program (SERDP) sponsored field tests to assess the skill of the both the underlying process models and the Bayesian Network implementation of UnMES. The process models are assessed by deterministic time-series comparisons and traditional metrics such as the coefficient of correlation. The summary statistic for the comparisons of burial results was r2 = 0.78, indicating that the model accounted for over 3/4 of the observed burial behavior. The Bayesian Network was evaluated by comparison of field observation histograms with predicted probability distributions using a Ranked Probability Skill Score. The models and expert system predictions generally agreed with the observations, providing substantial guidance regarding the munitions behavior. The spread in the output distributions predicted by UnMES correctly captured the observed variability.

Back to Top

Benefits

This Demonstration Version of UnMES brings together in a coherent and organized manner current knowledge on the manner in which munitions on the seabed can bury or migrate. The probabilistic construct feeds naturally into risk-assessment models used by site managers for remedial investigation decisions. With the expert system guidance regarding the timing, location and operational choices for assessment surveys and subsequent clean-up, these efforts can be more efficiently planned and executed, resulting in costs savings for the Department of Defense. Prediction of burial, which affects detection and classification performance by geophysical, acoustic and optical sensors, will guide optimal selection of sensor technologies. Knowledge of migration thresholds at remediation sites will allow evaluation of potential munitions relocation by storms of varying magnitudes. Additional benefit provided by the expert system is that it can function as a documented archive synthesizing records of laboratory and field research as well as databases of environmental conditions.

Back to Top

Points of Contact

Principal Investigator

Dr. Sarah Rennie

Johns Hopkins University Applied Physics Laboratory

Phone: 443-778-8178

Program Manager

Munitions Response

SERDP and ESTCP

Share