Causal reasoning for human supervised process reconfiguration

As safety is becoming an essential concern in industrial automation, an emerging area in automatic control is fault tolerant control. Within the various techniques, reconfiguration employs the redundancy in the plant and its control to make intelligent software that monitors the behavior of the whole. This paper analyses the reconfiguration problem from the point of view of large-scale processes under the responsibility of human operators. As analytical models cannot be envisaged for representing a process with hundred variables and several unpredictable operating modes, reconfiguration is proposed to rely on a simple qualitative model. This model represents the cause-effect relations between variables under the form of a directed graph. The graph is backward searched off line to find the action means on a variable. Then the most relevant remedial actions are selected online after a fault is detected. An example of a nuclear process is used to illustrate the method.