Network-Based Truth Maintenance System for Tactical Situation Assessment

Period of Performance: 12/22/2000 - 10/30/2001


Phase 1 SBIR

Recipient Firm

Charles River Analytics, Inc.
625 Mount Auburn Street Array
Cambridge, MA 02138
Principal Investigator


The growing digitization of the battlefield gives the intelligence analyst a unique opportunity to access large amounts of information collected over time across a variety of sensors to achieve an unparalleled level of tactical situation awareness. However, before using this array of dynamically changing tactical information, the data must be correlated and fused, and, most of all, managed in a truth maintenance system (TMS) ensuring logical data consistency. Rather than adopting a highly inefficient logic-based theorem-proving approach to maintain consistency across the entire database, we propose a Bayesian belief network (BN) approach that focuses truth maintenance only on the portions of the fused data relevant to the current assessment task. Each BN is constructed to assess a specific high-level situation in the form of the commander's priority intelligence requirement (PIR). Before posting incoming evidence at a BN node, a truth maintenance procedure is invoked to detect information inconsistency between the node's current state and the state of the evidence to be posted. In the case of inconsistency, the truth maintenance procedure isolates only relevant inconsistent nodes based on the causal dependency of the network. The proposed network-based TMS thus incrementally maintains only consistent BN states to ensure trustworthy situation assessment.Commercial applications of the proposed approach to truth maintenance in situation assessment incorporating Bayesian belief network technology exist in many areas including operation centers for complex process control (e.g., nuclear power plants), financial services, credit verification, loan approval, and rail and air traffic management centers. A belief network based situation assessment procedure that focuses only on the relevant data can also solve the information overload problem in high-value complex operational environments.