An Architecture to Support Incremental Automation of Complex Systems

Operators and domain practitioners often complain that automation is brittle, opaque, and ‘not worth the effort’ to use. This paper reviews automation problems and methods for the design of ‘cognitive automation.’ Cognitive automation is software intended to automate cognitive activities, such as situation assessment, monitoring, and fault management, that are currently performed by human operators. Limitations of current knowledge engineering methods—the key to robust cognitive automation—are presented. With this background, a design methodology and automation concept—incremental automation—are proposed. Incremental automation is software, which by design, serves as a cognitive apprentice to the operations staff of a complex dynamic system. Over time, as operations personnel refine and extend it, incremental automation accumulates knowledge that covers a broad range of operational experience. Furthermore, and again by design, the structure and processing used by incremental automation closely emulates structures and processes used by expert operators, thus facilitating software that is easy for domain practitioners, including operators, system designers, and management, to understand, repair, and enhance. This paper concludes with a description of an architecture to support incremental automation and its application in a NASA satellite ground control system. 1. THE INEVITABLE PROLIFERATION OF AUTOMATION Increasing levels of automation may help address the needs of modern systems and the organizations that manage them. Due to industrial competitiveness and budget constraints, organizations face increased pressures to reduce operations costs while increasing operating effectiveness. Though small when considered as a part of total life-cycle costs, operations costs are a substantial part of the annual operating budget for many systems. Thus, once a system has been developed and fielded, reducing staff via automation tools and functions may be an effective method to reduce operating costs. 1 NASA satellite ground control provides a telling example. Scientific spacecraft cost millions of dollars to build and launch. Once in orbit, however, their annual budgets are devoted almost entirely to the staff that manages the spacecraft and processes science data. Because of the large sunk cost to develop and launch the spacecraft, NASA and the science community are very reluctant to turn off spacecraft when their initial missions are complete. Budget reductions, however, force many missions to reduce either their operations budget by 70% or turn off the spacecraft. Additional motivation for increased automation comes in domains in which operating constraints (primarily the reduction in operating costs) are driving the introduction of increasingly automated control systems with a near-term goal of ‘lights-out’ operations. Lights-out automation is a term coined in the 1980’s in the area of manufacturing operations (e.g., Jaikumar, 1986) to describe a fully automated manufacturing facility in which robots, intelligent work cells, automated material handling systems, and computer-based controllers produced goods without human supervision or intervention (e.g., Shaiken, 1985; Warnecke & Steinhilper, 1985). Other domains including process control, telecommunications, and aerospace systems are now beginning to discuss ‘lights-out operations’ in which human operators will not be in attendance during certain shifts and automated systems will execute pre-planned activities, respond to anticipated anomalies, and report on the success or failure of operations after the fact. Together, these factors identify the need for increased and effective computer-based systems in operations, particularly automation of previously manual operations. To date, experience has had mixed results. Coupled with the everincreasing capabilities of modern computer technology, these factors and experiences suggest a need for operations automation that is robust, flexible, and easily extensible by domain practitioners. 2. PROBLEMS WITH AUTOMATION The introduction of any new technology carries with it associated benefits and costs. Unfortunately, rather than transforming system control into a more manageable function as intended, increasing levels of automation have too often resulted in both system performance and human interaction problems (see for example, Billings, 1997; Parasuraman & Mouloua, 1996; and Zsambok & Klein, 1997). There is some agreement, however, that the problem is not automation per se, but rather the inappropriate application and improper design of automation (Bainbridge, 1990; Mitchell & Sundström, 1997; Norman, 1990). Problems such as opacity, brittleness, and limited functionality are more often a result of poor automation design than the decision to incorporate automation into the system. As automation efforts turn increasingly from ‘control’ automation to ‘cognitive’ automation (Bainbridge, 1990), it becomes even more important to address the problems associated with automation design. Many of the problems with automation can be attributed to the technology-driven design process and the failure to adopt a systems-level perspective during the automation design