Incorporating product-line engineering techniques into agent-oriented software engineering for efficiently building safety-critical, multi-agent systems

Dehlinger, Joshua
Major Professor
Robyn R. Lutz
Committee Member
Journal Title
Journal ISSN
Volume Title
Research Projects
Organizational Units
Computer Science
Organizational Unit
Journal Issue
Computer Science

Safety-critical, agent-based systems are being developed without mechanisms and analysis techniques to discover, analyze and verify software requirements and prevent potential hazards. Agent-oriented, software-based approaches have provided powerful and natural high-level abstractions in which software developers can understand, model and develop complex, distributed systems. Yet, the realization of agent-oriented software development partially depends upon whether agent-based software systems can achieve reductions in development time and cost similar to other reuse-conscious software development methods. Further, agent-oriented software engineering (AOSE) currently does not adequately address: (1) requirements (specification) reuse in a way that is amenable to the reduction of the development cost by utilizing reusable assets, and (2) analysis techniques to evaluate safety.;This dissertation offers our AOSE methodology, Gaia-PL (Gaia-Product Line) for open, agent-based distributed software systems to capture requirements specifications that can be easily reused. Our methodology uses a product-line perspective to promote reuse in agent-based, software systems early in the development lifecycle so that software assets can be reused throughout the development lifecycle and system evolution.;The main contribution of this work is a requirements specification pattern that captures the dynamically changing design configurations of agents. Reuse is achieved by adopting a product-line approach into AOSE. Requirements specifications reuse is the ability to easily use previously defined requirements specifications from an earlier system and apply them to a new, slightly different system. This can significantly reduce the development time and cost of building an agent-based system.;For safety-critical agent-based systems, this dissertation incorporates reuse-oriented safety analysis methods for AOSE to allow the discovery of new safety requirements and the verification that the design satisfies the safety requirements. Specifically, Product-Line Software Fault Tree Analysis (PL-SFTA) and its automated tool, PLFaultCAT (Product- Line Fault Tree Creation and Analysis Tool), have been created to provide the technique and tool support for the safety analysis of safety-critical software product lines. The PL-SFTA allows for the identification of new safety requirements and the analysis of safety-critical requirements and requirement interactions. An AOSE-adapted Software Failure Modes, Effects and Criticality Analysis (SFMECA) technique has been created to support the derivation of a safety analysis asset using the specifications of Gaia-PL allowing for the identification of possible hazard scenarios and the failure points of specific agent roles. Using the assets generated via PL-SFTA and SFMECA, Bi-Directional Safety Analysis (BDSA) is shown to aid in the completeness of PL-SFTA and SFMECA, help verify the safety properties and strengthen the safety case when safety compliance to safety standards of the multi-agent system is necessary.;Results from an application to a large, safety-critical, multi-agent system product-line show that Gaia-PL provides strong reuse capabilities. Evaluation of the Gaia-PL methodology used in conjunction with the PL-SFTA, SFMECA and BDSA safety analysis techniques shows that safety analysis of an agent-based software system is feasible, reusable and efficient.