Cognitive modeling and dynamic probabilistic simulation of operating crew response to complex system accidents: Part 1: Overview of the IDAC Model

https://doi.org/10.1016/j.ress.2006.05.014Get rights and content

Abstract

This is the first in a series of five papers that discuss the information, decision, and action in crew context (IDAC) model for human reliability analysis (HRA). An example application of this modeling technique is also discussed in this series. The model is developed to probabilistically predict the responses of the nuclear power plant control room-operating crew during an accident for use in probabilistic risk assessments. The operator response spectrum includes cognitive, emotional, and physical activities during the course of the accident. This paper provides an overview of the IDAC architecture and principles of implementation as a HRA model. IDAC includes a crew model of three types of operators: decision maker, action taker, and consultant. Within the crew context, each individual operator's behaviors are simulated through a cognitive model under the influence of a number of explicitly modeled performance-influencing factors.

Introduction

This is the first in a series of five papers [1], [2], [3], [4] describing the IDAC model for human reliability analysis (HRA). The model has been developed to probabilistically predict the responses of a nuclear power plant (NPP) control room-operating crew-facing system anomalies for use in dynamic probabilistic risk assessment (DPRA). IDAC models various dimensions of operators’ responses including cognitive, emotional, and physical activities in the process of mitigating the consequence or/and bringing the system a safe state during accidents.

Over the past 20 years, the HRA discipline in relation to probabilistic risk assessments (PRA) has gone through an evolutionary and proliferating process. A number of proposed HRA methods and variants have been used in various PRAs. We find Reason's [5] three-level classification of human error as behavioral, contextual, and conceptual helpful as a basis for classification of HRAs methods. These three levels correspond approximately to the “what”, “where”, and “how” of the human response. At the behavioral level, errors are “classified according to some easily observable features of the erroneous behavior.” The contextual level “goes beyond the formal error characteristics and include limited assumptions about causality. In most cases, these assumptions do not stray far from the ‘surface’ data”. The conceptual level “is predicated on assumptions about the cognitive mechanisms involved in error production.” HRA methods can also be essentially characterized along the same lines.

The “first generation” HRA methods are mostly behavioral approaches, carrying at times some attributes of contextual approaches. The “second generation” HRA methods aspire to be of conceptual type. However, reaching such goal requires a predictive model with theoretical foundation and experimental validation. This is the aim of the IDAC model presented in this work.

IDAC model is originated from the IDA concept [6], [7], with many significant expansions and greater detail in concepts, modeling elements, procedures, and implementation [8], [9]. The IDAC operator behavior model was developed based on numerous relevant findings from cognitive psychology, behavioral science, neuroscience, human factors, social science, field observations, and various first- and second- generation HRA methodologies.

The IDAC presented here and four companion papers [1], [2], [3], [4] is best characterized as a simulation-based HRA. Examples of other similar efforts include cognitive environment simulation (CES) [10], operator-plant simulation model (OPSIM) [11], and man–machine integration design and analysis system (MIDAS) (e.g., [12]). CES uses artifical intelligence techniques to model a NPP control room crew responding to plant accidents. The objective in CES simulation is to predict the most likely crew response based on linking various segments of a data base representing operator knowledge. The content of the knowledge base and relations among its elements are explicitly set by the analyst prior to a simulation. OPSIM models operator cognitive behavior and identifies potential decision-based errors in the context of following procedural instructions. The probabilities of deviating from procedure instructions are not modeled in OPSIM. MIDAS was mainly developed for system designers to assess the ergonomic effects on operator performance for optimization new designs and rapid prototyping. The operator visual information perception, ergonomics of the system, and workplace design are the main aspects modeled in MIDAS.

The accident dynamics simulator (ADS) with integrated IDAC model (ADS–IDAC [13], [14], [15]) is a computer implementation of IDAC for NPP accidents which differs from the above approaches in the following key aspects: (1) IDAC at its core has a set of generic causal models which enable a relatively small set of rules and causal links to simulate the crew response to a vast array of specific situational contexts; and (2) within its modeling scope, ADS–IDAC considers a spectrum of possible operator responses and corresponding probabilities at each decision/action opportunity emerging in the simulation.

This (the first) paper provides an overview of the IDAC architecture and principles of its implementation as a HRA model in the ADS code for DPRA. The scopes of IDAC and ADS are shown in Fig. 1. As shown, IDAC models the individual and group behavior of the operating crew. Currently, three generic types of operator are modeled: decision maker, action taker, and consultant, each with somewhat different roles and responsibilities. These operators respond to system anomalies and interact with each other and with the system according to the group norm. The ADS code simulates accident scenarios and generates information about the external world (i.e., system state and environmental variables). This information is then used as input to the crew model (i.e., IDAC), which in turn simulates various types of operator response including actions on the system. Each individual operator model includes elements of the IDAC cognitive architecture (e.g., performance-influencing factors (PIFs), memory architecture) and model of cognitive process (e.g., information-processing model). The manner in which the IDAC model can be used as an HRA model within the DPRA framework for predictive analysis is also discussed in this paper.

Description and technical justifications of several important aspects of the IDAC model are delegated to the additional papers in this series. Paper 2 [1] describes the IDAC set of PIFs as well as their inter-dependencies and quantification method. Paper 2 also discusses the supporting evidence for the selection and organization of the influence paths among the PIFs. Also provided are details of inter-dependencies (expressed in the form of matrices and equations) among the external, internal, static, and dynamic PIFs. Further, we discuss appropriate ways of assessing the value or state of individual PIFs directly, or as a function of other PIFs.

Paper 3 [4] discusses various types of operator response to a situation. An operator's problem-solving process is divided into three types: information pre-processing (I), diagnosis and decision making (D), and action execution (A). Explicit and context-dependent behavior rules (i.e., rules-of-behavior in IDAC's terminology) regulate the process and activities of each type of response are developed. These rules are expressed in the form of tables and equations. The rules-of-behavior is a set of rules that govern all types of operator response (e.g., information perception, cognitive and emotional responses). The rules take physical and psychological factors as input and generate behavior as output. The rules reflect the effects of memory, knowledge, and emotions, together with the core cognitive and intellectual faculties. Discussions of the rules-of-behavior applied to different cognitive process are overviewed in this paper (Paper 1) and more detailed discussion is in Paper 3 [4].

Paper 4 [2] establishes the connection between the PIFs (discussed in Paper 2 [1]) and the rules-of-behavior. Literature support and justifications are provided for the assessment of PIFs’ influence on operator responses. Paper 4 also provides a simple approach to calculating normalized probabilities of alternative behaviors within the above spectrum.

Paper 5 [3] provides a detailed example of implementing IDAC in the ADS code to demonstrate the practicality of integrating a detailed cognitive HRA model within a DPRA framework. A simple plant system model and a simplified version of IDAC [9] are used for the demonstration to save computational time. Current research effort combines the full-scale IDAC model with a RELAP5 plant model [16] for a more realistic simulation of NPP accident scenarios.

The current paper is organized as follows. An overview of the IDAC model is discussed in Section 2. The dynamics of an operator interacting with the external world (i.e., the crew, system, and environment) and of the operator's inner mind are discussed as well as the principles for calculating response probabilities. Section 3 is an overview of the problem-solving and decision-making elements of IDAC's model of a single operator. Section 4 describes the key aspects of the process of problem solving and decision making, i.e., how the model simulates operator's response. Section 5 provides an overview of the integrated ADS–IDAC simulation framework. Section 6 shows how human error and causes can be identified with use of ADS–IDAC. Section 7 offers concluding remarks including current status and a roadmap for the future.

Section snippets

Overview of the IDAC model

This section provides an overview of the IDAC modeling on the dynamics of an operator interaction with the external world (i.e., the system, environment, and other crew members) and his cognitive and emotional response to a situation. The stochastic aspect and IDAC's approach to calculating response probabilities are also summarized.

IDAC diagnosis and decision-making model elements

Studies [26] suggest that people use two types of domain-specific knowledge to solve problems: conceptual knowledge and procedural knowledge. “Conceptual knowledge is knowledge of the principles of the domain,” and “procedural knowledge is knowledge of how to carry out operations, that is, mechanisms of problem solving.” The representation model of conceptual knowledge is discussed in Section 3.2.3. Procedural knowledge is modeled by the IDAC cognitive process. The cognitive elements are key

Overview of dynamic process for problem solving

The combination of cognitive processes and observable actions of an operator during the course of an accident is a continuum. The entire process may be divided into smaller phases in terms of dominant goals or modes of response such as situation assessment, searching for the cause, and selection and execution of the response and recovery plan. Each of these phases can be further divided into sub-phases (e.g., following specific segments of a procedure) with specific and distinct cognitive and

Use of IDAC in dynamic simulation environment

Due to the variety, quantity, and detail of the input information, as well as complexity of applying its internal rules, the IDAC model is best implemented through a computer simulation such as the ADS simulation environment [14], [55]. The most recent release of ADS program contains six modules, as shown in Fig. 13. User interface module enables the user to edit the inputs, such as initial conditions of the system and operator models and to control the analysis parameters. Scheduler module

Errors from IDAC perspective

The human error literature offers a variety of classifications of error. Examples are:

  • intentional and unintentional errors [50],

  • errors of omission and commission [50],

  • skill-, rule-, and knowledge-based errors [5],

  • slip, lapse, mistake, and violation [5],

  • genotype and phenotype errors [58],

  • active and latent errors [59],

  • internal mode of malfunction and external mode of human malfunction [34] (or endogenous and exogenous errors [60]),

  • information perception, decision-making, and action execution

Concluding remarks

IDAC represents a sustained effort over the past 15 years for developing a causal model for operators’ behavior deep into the cognitive level. The series of five papers in which the present one offers the overview and fundamentals, document the more recent progress. The work on IDAC has accompanied the development of the dynamic PRA code, ADS, which has been used as a vehicle to test IDAC and the feasibility of integrated dynamic PRA for NPP risk applications. The evolution of IDAC and ADS has

Acknowledgements

The development of IDAC and ADS have been in part supported by the Electric Power Research Institute and the US Nuclear Regulatory Commission, Office of Research under a collaborative research agreement with the University of Maryland Center for Risk and Reliability. Several visiting scientists at the University of Maryland from the National Maritime Research Institute of Japan have made significant contributions to the development of ADS and implementation of IDAC. Current work on IDAC and ADS

References (65)

  • Chang YHJ, Mosleh A. Cognitive modeling and dynamic probabilistic simulation of operating crew response to complex...
  • Chang YHJ, Mosleh A. Cognitive modeling and dynamic probabilistic simulation of operating crew response to complex...
  • Chang YHJ, Mosleh A. Cognitive modeling and dynamic probabilistic simulation of operating crew response to complex...
  • Chang YHJ, Mosleh A. Cognitive modeling and dynamic probabilistic simulation of operating crew response to complex...
  • J. Reason

    Human error

    (1990)
  • S.-H. Shen

    IDA: a cognitive model of nuclear power plant operator response during abnormal conditions

    (1994)
  • Chang YH, Mosleh A. Cognitive modeling and dynamic probabilistic simulation of operating crew response to complex...
  • Woods DD, Roth EM, Pople Jr HE, Embrey D. Cognitive environment simulation: an artificial intelligence system for human...
  • V.N. Dang

    Modeling operator cognition for accident sequence analysis: development of an operator-plant simulation, in Nuclear Engineering Department

    (1996)
  • Corker K, Pisanich G, Bunzo M. A cognitive system model for human/automation dynamics in airspace management, 1997....
  • Chang YH, Mosleh A. Dynamic PRA using ADS with RELAP5 code as its thermal hydraulic module. In: Probabilistic safety...
  • Chang YH, Mosleh A. ADS: a computer program for dynamic probabilistic risk assessment. In: International topical...
  • Chang YH, Mosleh A, Dang VN. Dynamic probabilistic risk assessment-framework, tool, and application. In: Society of...
  • K.E. Carlson et al.

    RELAP5/MOD3 code manual (Draft)

    (1990)
  • C.R. Paris et al.

    Teamwork in multi-person systems: a review and analysis

    Ergonomics

    (2000)
  • F.V. Jensen

    Bayesian networks and decision graphs

    (2001)
  • J. Pearl

    Probabilistic reasoning in intelligent systems: networks of plausible inference

    (2001)
  • E. Salas et al.

    Toward an understanding of team performance and training, in teams: their training and performance

  • K.A. Smith-Jentsch et al.

    Team dimentional training: a strategy for guided team self-correction, in making decision under stress

  • J.A. Connon-Bowers et al.

    Reflections on shared cognition

    J Org Behav

    (2001)
  • M.T. Brannick et al.

    An overview of team performance measurement

  • US Nuclear Regulatory Commission. Onsite analysis of the human factors of an event at LaSalle 2 April 20, (reactor...
  • Cited by (0)

    View full text