[Paper Review] Multi-Agent Optimization for Safety Analysis of Cyber-Physical Systems: Position Paper

Önder Gürcan, Nataliya Yakymets+2 more|arXiv (Cornell University)|Mar 25, 2024

Smart Grid Security and Resilience4 citations

TL;DR

This paper proposes a multi-agent optimization framework based on Adaptive Multi-Agent Systems (AMAS) to automate trade-off decisions in Failure Mode, Effects, and Criticality Analysis (FMECA) for complex cyber-physical systems (CPS). By modeling failure modes and preventive actions as agents that cooperatively select optimal configurations under cost and criticality constraints, the method reduces manual effort and resource overuse while ensuring safety-criticality remains below thresholds.

ABSTRACT

Failure Mode, Effects and Criticality Analysis (FMECA) is one of the safety analysis methods recommended by most of the international standards. The classical FMECA is made in a form of a table filled in either manually or by using safety analysis tools. In both cases, the design engineers have to choose the trade-offs between safety and other development constraints. In the case of complex cyber-physical systems (CPS) with thousands of specified constraints, this may lead to severe problems and significantly impact the overall criticality of CPS. In this paper, we propose to adopt optimization techniques to automate the decision making process conducted after FMECA of CPS. We describe a multi-agent based optimization method which extends classical FMECA for offering optimal solutions in terms of criticality and development constraints of CPS.

Motivation & Objective

• To address the growing complexity and manual burden of classical FMECA in large-scale cyber-physical systems (CPS) with thousands of failure modes and preventive actions.
• To automate the selection of optimal preventive actions that balance safety-criticality, cost, and development constraints.
• To overcome limitations in existing FMECA tools that cannot detect redundant or overlapping preventive actions, leading to resource overuse.
• To leverage self-organizing multi-agent systems for scalable, distributed optimization of safety-critical system configurations.
• To lay the foundation for future integration of behavioral models and runtime constraints into automated safety analysis.

Proposed method

• Models failure modes and recommended preventive actions as autonomous agents within an Adaptive Multi-Agent System (AMAS) framework.
• Uses a criticality function $ c_q(t) = \frac{1}{n_i(t)} $, where $ n_i(t) $ is the number of failure modes for component $ p_i $, to quantify agent criticality.
• Employs cooperative behaviors via feedback mechanisms: 'select more' ($ \uparrow $), 'select less' ($ \downarrow $), and 'selection good' ($ \approx $) to guide action selection.
• Implements reorganization behaviors using add(\{g,p\}) and remove(\{g,p\}) actions to form or suppress selection relations between failure modes and preventive actions.
• Enables agents to make decentralized decisions: a preventive action agent only acts if it is the most critical among its neighbors, ensuring load balancing and avoiding redundant selections.
• Applies a distributed, self-organizing approach to optimize system-wide criticality while respecting predefined constraints such as cost and resource limits.

Figure 1: A simple example of FMEA and FMECA.

Experimental results

Research questions

• RQ1 How can the manual and error-prone process of selecting preventive actions in FMECA be automated for large-scale CPS?
• RQ2 What multi-agent mechanism enables optimal trade-offs between safety-criticality and development constraints like cost and time?
• RQ3 How can overlapping or redundant preventive actions be detected and avoided in FMECA using agent-based coordination?
• RQ4 In what way does the AMAS framework improve scalability and robustness in safety-critical system configuration optimization?
• RQ5 What extensions are needed to support timing constraints and behavioral models in future FMECA automation?

Key findings

• The AMAS-based approach successfully models failure modes and preventive actions as agents that cooperatively select optimal configurations, reducing reliance on manual decision-making.
• The use of a criticality-based agent function $ c_q(t) = \frac{1}{n_i(t)} $ enables dynamic, decentralized assessment of component criticality.
• Cooperative feedback mechanisms ('select more', 'select less', 'selection good') allow agents to self-organize and avoid redundant action selection.
• The model demonstrates scalability through distributed decision-making, where only the most critical agent among neighbors acts, preventing resource overuse.
• The framework provides a foundation for future integration of behavioral models and timing constraints to support fully automated, run-time-aware FMECA.
• The approach addresses key limitations of classical FMECA by detecting cases where multiple failures can be resolved by a single action or where alternatives exist, thus optimizing resource allocation.

Figure 2: The automated FMEA process.