Computer Simulations for Fault Diagnosis Training: From Simulation to Live System Performance

Abstract

Three methods of training for fault diagnosis were studied: traditional instruction, context-specific computer simulations, and context-free simulations. Each of thirty-six powerplant maintenance trainees were trained with one of these methods, and then, transferred to real aircraft powerplants and, later, to simulations of an auto-pilot system. For the aircraft powerplants, it was found that traditional instruction was only superior if explicit demonstrations were provided for the exact failures to be encountered. With respect to overall performance, context-free training did not produce improved performance on the unfamiliar auto-pilot system. Considering correlates of live system performance, it was found that performance in the context-free simulation was a good predictor, while performance in the context-specific simulations was by no means useful. Trainee Grade Point Averages and scores on traditional quizzes were also highly correlated with live system performance.