Intelligent Tutoring Systems for Aviation Automation
CHRISTINE M. MITCHELL, Alan R. Chappell, Wm. Michael Gray, Alex Quinn, David A. Thurman · SAE technical papers on CD-ROM/SAE technical paper series · 2000
This paper begins with a discussion of a cognitive engineering model, the operator function model (OFM), to guide design of artifacts for human interaction with complex systems. Such artifacts include operator aids, associates, and tutors. The paper presents an overview of the current implementation and evolution of the operator function model (OFM) and OFMspert, its computational implementation. It describes how OFMspert has been extended to support the design of two intelligent tutoring systems (ITS) for operational control of safety-critical systems. Proof-of-concept demonstrations and evaluation teaching MD-11 transition pilots vertical navigation and a case-based tutor teaching currently certified MD-11 pilots new procedures, adapted since their certification training. 1. Background Almost two decades ago, Mitchell defined the operator function model (OFM) (Mitchell, 1987). The OFM is both a cognitive engineering and human-machine systems engineering model. It was created to provide a mathematical and visual representation of operator activities in control of complex, dynamic systems. The OFM makes explicit assumptions about modeling operator behavior in complex systems. These include hierarchy, heterarchy, and non-determinism. The OFM and its properties are extensively described in various publications (Mitchell, 1999; Thurman, Chappell, & Mitchell, 1998b). Inspection of Figure 1 shows a portion of an OFM implemented for aircraft navigation. Pieces of an OFM are often called OFM trees or subtrees. The OFM is a static model that represents when and how trees become active. The tree depicted in Figure 1 is active because the initiator, aircraft is not within limits of assigned heading, is true. The top-level activity, turn to assigned heading, decomposes into subtrees. Decompositions can take a variety of forms. Activities can be sequential (SEQ)—order is important. Others are heterogeneous (AND)—all activities must be performed but no order is required. Others are choices: an OR decomposition allows the operator to execute one or more activities; whereas hi an XOR decomposition activities are mutually exclusive and the operate must select exactly one. Figure 1 shows the decomposition for the activity, set FCP (flight control panel) heading target,' decomposed to the action level. Note that activity nodes are depicted as rounded-corner rectangles. Actions, the lowest level activity, are an exception. Actions are denoted by square-cornered rectangles. This syntax makes it easier to watch the run-time system. OFMspert is a computational and dynamic form of the OFM. OFMspert displays those portions of the OFM that are active at the present time. In 1986, the OFMspert project began. The Nii (Stanford) blackboard (Nii, 1986a; Nii, 1986b), an artificial intelligence methodology, offered an ideal software architecture with which to implement the OFM as a run-time system. The Nii blackboard is hierarchical, heterarchical, dynamic, and, in real time, processes incoming data to support current hypotheses, add new hypotheses, or reduce the likelihood of existing hypotheses. ACTIN (actions interpreter) is OFMspert's blackboard, displaying active trees and, in real time, connecting detected actions to expected actions. Linking detected actions to expected actions and using blackboard knowledge sources, specialized functions or methods, to ensure that the activity is proceeding correctly defines the intelligence that OFMspert brings to its applications. Failure to link an action to an existing activity or failure to detect an expected action indicates possible user errors. Figure 2 depicts a representation of the current OFMspert architecture. It is both domain and application independent. Domain-specific information, that is, a detailed description of the system of interest such as satellite control, aircraft navigation, or electronics manufacturing, is defined in files. Application-specific information is similar. Files define if the system will be an intelligent tutor, an operator's associate, or control automation. OFMspert reads these files at initialization and customizes the generic OFMspert (c)2000 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization. to the domain and application of interest. Since the current version of OFMspert is implemented in standard Java, OFMspert is also platform independent. 2. Intelligent Tutoring Systems: Operational Training for Complex Systems In the workplace, training is mandatory. The rate of emerging, powerful, and inexpensive technology shows no signs of decreasing. Inevitably, some of this technology finds its way into most work domains. The rate of change in work domains due to the introduction of technology shows no sign of slowing. The introduction of new technology often fundamentally changes system operation (Woods, Johannesen, Cook, & Barter, 1994). Rapid change in the workplace means rapid change in the knowledge and operational skills to manage the changing systems. Thus, training becomes very important. Particularly for complex, often safety-critical, systems, effective and timely training, which keeps experienced operators abreast of how the system has changed and new methods for managing it, is mandatory to ensure system safety and efficiency. Legend push POP HDOnRK