DETECTION OF SYSTEM FAILURE BY HUMAN OPERATOR — MATHEMATICAL MODEL AND EXPERIMENTS

Abstract This paper proposes a model of human operators' failure detection behavior in actively controlling a linear system and passively monitoring signals on the display screen. Estimation and signal detection theories were appliied to the development of this model. Its mechanism of failure detection is expressed by Wald's Sequential Probability Ratio Test, and two autoregressive models are used to predict the values at one step ahead of signals - signals on the display screen and a displacement of the manipulator - from their current values as perceived by the human operator. One of these autoregressive models is for hypothesis H 0 where the system is in normal working order, and the other is for hypothesis H 1 where the system is out of order. A major feature of the proposed model is that the information accessible to the controller includes proprioceptive information associated with control manipulation in addition to visual information. The detection time and accuracy predicted by this model corresponded well with the data obtained from experiments over a wide range for the controller and monitor.