Setting the stage

The human operators of complex systems, such as computer-integrated manufacturing systems, power plants, and aircraft flight decks, are highly educated and trained. They can monitor and manage their particular systems under normal conditions and during system malfunctions. These operators typically must deal with an enormous quantity of data conveying system health that is presented to them on multiple computer screens.1,2 To assist with this data avalanche, automation’s role in the control of these complex systems has expanded. The increase of automation changes the operator’s role from a controller to a monitor and, when needed, a troubleshooter responsible for fault detection, diagnosis, and compensation.3 Although automation has shifted operator responsibilities, there is no substitute for human decision making and experience to set high-level system goals, monitor system states, and compensate for anomalies that the automation systems were not designed to handle.4 When anomalies occur for which automation cannot compensate, these operators must quickly diagnose and correct the anomalies. They identify faulty system components by observing incorrect systems states, creating hypotheses based on their experience, and testing each hypothesis by manipulating the system. The interactions between, dynamics of, and sheer number of elements in these systems complicate this method of anomaly diagnosis and compensation. The huge number of system components increases the alternatives available to explain system anomalies. Furthermore, determining which anomalies are causes and which are effects is difficult because any single anomaly quickly proliferates through the systems to cause other anomalies. Steam-propulsion plants are prototypical examples of complex engineering systems. Engineers have studied them extensively and developed simulations of them to assist training about and the control of these complex engineering systems. Steampropulsion systems deconstruct into subsystems such as fuel-oil, lubrication-oil, main-steam, auxiliary-steam, cooling, and turbine. Likewise, each subsystem (the main-steam system, for example) decomposes into subsubsystems, such as air supply, economizer, and boiler. Finally, these subsubsystems are composed of physical components, which form physical interconnections to create a network of interactions and failure points. This decomposition of the system parallels the operators’ knowledge. Generally, plant operators and system designers think of higher-level pieces of the system in hierarchical, functional terms, whereas they understand lower levels in terms of physical interconnections. For example, they think of the main-steam subsystem as a heat-transfer unit, but they consider the boiler as an interconnected collection of physical components.