Human Decision-Making in Computer Aided Fault Diagnosis

Abstract

Abstract : This report summarizes a series of six experiments conducted to increase our understanding of human performance on diagnostic tasks, and in the process to investigate the feasibility of using context-free computer-based simulations to train troubleshooting skills. Three simulated diagnostic tasks were developed: a simple context-free task, a complex context-free task, and a context-specific task (simulation of aircraft powerplants). The six experiments evaluated the effects of computer aiding on performance of each task and on subsequent unaided performance, using different task mixes, subjects (4 to 48 engineering or technical trainees), and conditions (self-pacing vs. forced pacing; feedback loops). Computer aiding reduced the number of tests required to diagnose simple problems and enhanced subsequent unaided performance except when subjects were under time pressures. Training on the simple task with computer aiding first inhibited and then enhanced performance on the complex context-free task. Training on the context-free tasks improved performance on the context- specific task. Results provide a data base for both theoretical issues in fault diagnosis and practical application of computer aiding to live system performance.