Young Investigator Program (8.5): Preventing Complex Failures of Human Interactive Systems with Erroneous Behavior Generation and Robust Human Task Behavior Patterns
Abstract
Failures in complex systems often occur because of unexpected erroneous human interactions. While conventional human factors engineering techniques have made strides towards reducing incidents of erroneous behavior and improving system safety, they are not exhaustive and can miss critical conditions. To address this, researchers have investigated how model checking (an automated proof method), task analytic models (models of human behavior), and erroneous behavior generation techniques can be used together to prove properties about the safety of human-interactive system models. Although powerful, the erroneous behavior generation techniques used with model checking are based on incompatible phenomenological and genotypical taxonomies. Further, model checking analyses are difficult to use and scale very badly, limiting their application. There is a real need for methods that will enable more complete model checking analyses that evaluate how erroneous human behavior can contribute to system failures and allow engineers to design safety-critical human procedures while avoiding the limitations of model checking. The project is investigating a new erroneous behavior taxonomy and associated erroneous behavior generation technique that can be used with model checking to discover patterns of human task behavior that are robust to erroneous human behavior. Further, we intend to show that these generic patterns can be used and reused by engineers to design human procedures that will prevent complex system failure while avoiding the scalability limitations of model checking. We plan to complete our objectives through the completion of three research goals. First, we will develop a new taxonomy of erroneous human behavior based on where in a task model a humanÕs behavior diverges. We will analytically show that this taxonomy will encompass the leading genotypical and phenomenological taxonomies. Second, we will use this new taxonomy as the basis for a novel, more complete erroneous behavior generation method. This will be incorporated into a model checking approach that will enable the contribution of potentially unanticipated erroneous behavior in complex system failures to be discovered. We will test this method to ensure it exhibits the desired behaviors and use case studies to evaluate its capacity to find unexpected system failures. Finally, we will show that the generation method and model checking analysis can be used to identify generic, reusable human task behavior patterns that are robust to erroneous acts. We will apply these patterns to variations of the examined case studies to demonstrate their ability to avoid complex failures. In pursuing each of these goals, developments will be applied to and evaluated using case studies from Army UAV operations and fire procedures. By connecting phenomenological and genotypical classifications, the new erroneous behavior taxonomy will give users additional insights into why classified acts occurred. The taxonomy should thus be invaluable in system and accident analyses. The taxonomy will also enable the development of the erroneous behavior generation method and its associated model checking analysis. This will give analysts an unprecedented ability to find unexpected ways that erroneous behavior can contribute to complex system failures. The generation process will further enable the discovery of the robust task patterns. The task patterns are significant because they will enable engineers to design human procedures that will avoid complex failures associated with erroneous human acts. Because these task patterns will be created and model checked by researchers, they will only need to be used by engineers. The size of the patterns will avoid scalability restrictions. Thus, the use of the patterns will give engineers the means to achieve the benefits of model checking while avoiding its usability and scalability limitations.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 12, 2017
- Source ID
- W911NF1510474
Entities
People
- Matthew L Bolton
Organizations
- Army Contracting Command
- United States Army
- University at Buffalo