Increasing Aviation Safety Using Human Performance Modeling Tools : An Air Man-machine Integration Design and Analysis System Application

Human Performance Modeling (HPM) tools are computational, human-out-of-the-loop (HOOTL) representations of several micro models of operatorenvironment performance used to predict complex humansystem interactions. HOOTL processes provide economical (in terms of time and money) means of studying complex human-system performance. As technologies and automation increase to assist the human operator in the increasingly cognitively demanding world, human-related vulnerabilities may arise that may impact the system safety by increasing procedural error rates. Hollnagel’s conceptualization of human error will be used as the theory behind a HOOTL simulation currently underway at NASA Ames Research Center to predict human error in the aviation environment in surface operations [1]. One of the HOOTL simulation tools being used to generate human-system performance predictions is the emergent HOOTL tool termed Air Man-machine Integration Design and Analysis System (MIDAS). This paper will outline the current understanding of factors underlying human error and the considerations that need to be heeded in developing HOOTL simulations for human-automation predictions. These HOOTL simulations will be shown to be effective means of predicting system vulnerabilities and will allude to possible intervention strategies. BACKGROUND AND INTRODUCTION Two methods exist for studying human performance in complex systems: Human-in-the-loop (HITL) high-fidelity simulations or computational humanout-of-the-loop (HOOTL) predictive simulations. The use of HITL simulation has been proposed as a methodology for examining human-systems performance in a safe and controlled environment in the surface transportation and aviation communities [2]. This technique has proven to be successful in accomplishing the goal of safely and realistically evaluating human-system behavior but has the disadvantage of being very complex and costly, often times prohibiting its use. In contrast, HOOTL simulations can be less expensive and used at an earlier process in the development of a product, system or technology. HOOTL simulation tools are computer-based simulation processes where human characteristics, taken from years of research from respective fields, are embedded within a computer software structure to represent the human operator interacting with computer-generated representations of the operating environment [2,3,4]. The human characteristics in many of the integrated HOOTL simulation tools include visual and auditory perceptual and attentional systems, anthropometric characteristics, and environmental characteristics (including workstations as well as the outside environment). These structures feed-forward and feedback with the goal of predicting human behavior. These complex integrated HOOTL simulation tools permit researchers to formulate procedures, generate and test hypotheses, identify variables for upcoming HITL simulations, and refine the procedures to ensure that they can be successfully completed in the time allotted for the given environmental demands. The output measures of interest for HOOTL simulation efforts from the aviation community generally include workload and timing measures. These measures have been validated across multiple domains: helicopter operations [5,6,7], nuclear power-plant control electronic list design for emergency operations [8], and advanced concepts in aviation [9,10]. Integrated human performance models (HPM) a form of HOOTL simulation, include procedural static models of human performance, anthropometric models of human performance, complex, dynamic representations of human performance, and cognitive performance [11]. A number of model representations are required to create the dynamic model representation. The dynamic representation of human performance requires the developer to create static representations of the overall task structure that is performed by the agent in the simulation. Since the human operator is simulated, the risks to the human operator and the costs associated with system experimentation are greatly reduced: no experimenters, no subjects, and no testing time. One criticism of HOOTL tools is that the software only predicts input-output behavior in mechanistic terms. Gore and Corker indicate that the integrated structure of the tools does more than solely represent input-output behavior [13]. The framework integrates many aspects of human performance allowing each micro model component to behave in its required method, the integration of which replicates a human 1 For further discussion of the advantages and disadvantages of HPM, please consult [2]. Cognitive Modeling Development One fundamental component within the integrated HPM is the cognitive component and many of the HPMs are now attempting to augment their cognitive representations due to increases in cognitive demands associated with recent advances in the operational environment. Cognitive modeling concepts were integrated into the engineering models’ philosophy in order to assist in predicting complex human operations. The overall philosophy behind the use of cognitive modeling was to provide engineering-based models of human performance. The engineering-based models of human performance permit a priori predictions of human behavior of a very restricted set of behaviors in response to specific tasks. Human performance modeling has traditionally been used to predict sensory processes [15], aspects of human cognition [16], and human motor responses to system tasks [3,17,18]. Human performance modeling tools are currently undergoing another developmental shift. The attempt now is for the HPM to be sensitive to situations that confront a virtual human in systems similar to the HITL situations. The growth in human performance modeling has been to examine human performance in systems including system monitoring (thereby taking information in from the environment) as opposed to the closed-loop view of the human as a mathematical relationship between input and output to a system. In fact, human-computer simulation modeling programs have been proposed to study human performance interacting with systems, and to support prediction of future system state [19]. These hybrids of continuous control, discrete control and critical decisionmaking models have been undertaken to represent the “internal models and cognitive function” of the human operator in complex control systems. These hybrid systems involve a critical coupling among humans and machines in a shifting and context sensitive function. The Man-machine Integration Design and Analysis System (MIDAS) is an example of such a hybrid tool that utilizes an emergent behavior approach to modeling an individual’s performance [2]. MIDAS’ general structure is made up of interconnected systems that interact with each other in a closed-loop fashion (Figure 1). Objects in MIDAS exchange messages through agent architectures. MIDAS’ agent architecture is made up of physical component agents and human operator agents [8]. Physical component agents can use commercially available computer-aided design (CAD) databases to graphically represent physical entities in an environment. Physical world agents are the external environments such as terrain and aeronautical equipment. The human operator agents are made up of human performance representations of cognitive, perceptual and motor operations of a task. These models describe within their limits of accuracy the responses that can be expected of the human operator. The attention demands are based on Wickens’ Multiple Resource Principle [20] and incorporate a task loading index created by McCracken-Aldrich for quantifying attention [21] along the visual, auditory, cognitive and psychomotor (VACP) resources. In addition, MIDAS possesses degradation functions that incorporate the effect of stressors on skill performance through Rasmussen’s skill-, ruleand knowledge-based decisions [22]. Symbolic Operator Model