The real-time collection and filtering of the enormous amounts of data resulting from training simulations involving hundreds of entities is a challenge for training system architects. Traditional approaches have relied on human observer/trainers (O/T) to tag key events and prepare the After Action Reviews (AAR), which identified what happened (particularly events contributing to metrics for mission success), why metrics were not met (precursor events that contributed to the metric events), how the critical sequences of events arose (identifying decision points), and provide timely learning points. For collective training simulations, the O/Ts are often overwhelmed in terms of tracking individual behaviors and skills. Automated approaches for capturing human performance data are a preferred method that can provide feedback to each member of a team or unit.

This paper describes the application of a novel use of XML Stylesheet Language Transformations to process simulation event stream using an Event-Condition-Action (ECA) rule engine. The human performance data capture used the following process:

1. Data Collection from multiple sources, including IEEE 1278 Distributed Interactive Simulation (DIS) data and other network data.

2. Protocol Filtering, which is done to reduce the processing workload in the later stages of the pipeline.

3. XML Conversion, where the filtered data streams are converted to a neutral XML format that allows processing using ECA rules.

4. Event Detection, which is done using XSLT to extract key events.

5. Event Processing, which is done to relate selected event sequences to human performance standards.

The Embedded Training Group at General Dynamics Land Systems has applied this approach to measure learner performance in distributed interactive simulations, including driver training and gunnery training. Ongoing work is related to using competency standards to specify the ECA rules and generate the appropriate XSLT.