FIXIT: An Architecture to Support Recognition-Primed Decision Making in Complex System Fault Management Activities

This research explores the use of the theory of recognition primed decision making and the technology of case-based reasoning to store fault management experience and make it available to operators confronting similar anomalous situations. Specifically, this project uses case-based reasoning technology to construct a knowledge base of actual fault management experience. This knowledge base is organized so as to enable operators to rapidly recognize a fault and retrieve information about previous or similar fault occurrences. The Fault Information Extraction and Investigation Tool (FIXIT), a case-based architecture for computationally encoding fault management experience, is described. FIXIT was implemented in proof-ofconcept form for satellite ground controllers. An empirical study with NASA controllers showed that fault management using FIXIT was consistently superior to fault detection and response using existing fault management resources of NASA satellite ground control centers. Introduction The need to better understand and model the process of fault management is motivated by three prevalent trends in the control of complex dynamic systems. The first results from the increasing levels of automation found in most modern control systems. Most modern automation handles nonnominal situations poorly. Human operators are frequently left to deal with anomalous and unexpected system conditions (Sheridan, 1992). This trend points to the importance of developing automated systems which can detect, diagnose, and respond to anomalies and support the human operator when the limits of automation are reached. A second trend relates to the changing role of the operator in human-machine systems. Reduced budgets and a corresponding desire to reduce the necessary number of operations personnel mean that operators are increasingly controlling multiple, independent systems concurrently (see Thurman & Mitchell, 1995). The result has been a marked reduction in the detailed knowledge operators have with which to respond to unexpected situations or to manage faults. The third trend results from high levels of turnover among operations personnel. In general, when personnel changes occur, relatively expert personnel are replaced by less experienced people. In order to judge whether or not a controlled process is in error, operators must have knowledge of the process being performed or expectations about its correct outcome (Hutchins, 1995). Operators’ ability to successfully perform fault management is directly related to their level of experience with the controlled system. As the amount of time and experience that operators have with a specific system decrease, the speed and efficiency of fault management also decreases. One of the results of giving operators responsibility to control multiple independent systems, coupled with high operator turnover, may be that operators do not have the relevant knowledge to successfully manage faults. Given the context in which fault management is conducted (i.e., high workload and high likelihood of distractions and interruptions) even if the operator possesses the relevant knowledge it is easy for a memory lapse to occur (Woods, Johannesen, Cook, & Sarter, 1994). The brittle nature of modern automation also means that operators may be called upon to perform without the aid of automation at the very time they are in most need of assistance (Woods et al., 1994). The Fault Information Extraction and Investigation Tool (FIXIT) is proposed as one strategy to help mitigate these problems by encoding previous fault management activities and making that information available to operations personnel. FIXIT stores and makes available to the operator past fault management experiences in such a way as to facilitate recognition of the current fault as an instance or variant of previous experience. Based on both the theory of recognitionprimed decision making and the technology of case-based reasoning, FIXIT is intended to embody the properties of human-centered automation--that is, automation that reflects how humans solve problems and make decisions. As such, it