The US military is in need of technical personnel capable of diagnosing and resolving operational system issues as part of their maintenance duties. Schoolhouse training needs to deliver both familiarity with fundamental principles and readiness to maintain specific devices. The usefulness of the training after deployment is challenged by skill decay, systems updates, and idiosyncratic systems that behave differently from a general model because of usage or wear.

A digital twin provides a simulation that models one individual device, rather than a general model of an idealized device. The increasing use of digital twins, especially in industrial applications, provides a wealth of data that can create training specific to one device. Training with a digital twin can create a high-fidelity experience to accelerate learning, minimize skill decay, improve transfer of skills from the schoolhouse to the operational platform, and support on-the-job training once deployed. 

We address two of the challenges in using digital twins for training. First, we describe an approach to train underlying principles with a model of physical system performance across different systems (e.g., propulsion, sonar, radar, fire control), subsystems (e.g., mechanical, electrical) and components (e.g., valves, actuators). Second, we describe how to author training with the model that enables focusing on key parts of a process, accepting learner input, predicting device outputs, and assessing learner performance for feedback to learners or instructors.

The approach to turn data into training relies on machine learning from the behavior of a single device. We demonstrate the accuracy of our machine learning with data from a commercial jet engine. We show that our approach predicts how an individual engine will respond to wear and maintenance. As a result, the digital twin can present a number of training scenarios with automated instructional feedback.