Continuous Resilience in Cyber-Physical Systems of Systems: Extending Architectural Models through Adaptive Coordination and Learning

TL;DR

This paper introduces a novel architectural framework for cyber-physical systems of systems that enables continuous, adaptive resilience through real-time coordination and learning layers, moving beyond static resilience models.

Contribution

It proposes an Adaptive Coordination Layer and a redefined Learning Layer to operationalize resilience as a dynamic, data-driven process integrating short-term response and long-term learning.

Findings

Implementation variants include rule-based, KPI-driven, and AI-supported mechanisms.

The architecture enables real-time risk detection and dynamic countermeasure coordination.

It provides a methodological basis for next-generation adaptive resilient CPSoS.

Abstract

Cyber-physical systems of systems (CPSoS) are highly complex, dynamic environments in which technical, cybernetic and organisational subsystems interact closely with one another. Dynamic, continuously adaptable resilience is required to ensure their functionality under variable conditions. However, existing resilience architectures usually only deal with adaptation implicitly and thus remain predominantly static. This paper addresses this gap by introducing a new Adaptive Coordination Layer (ACL) and conceptually redefining the Adaptation & Learning Layer (AL). The ACL acts as an operational control layer that detects risks in real time, prioritises countermeasures and coordinates them dynamically. The AL is reinterpreted as a strategic-cooperative layer that evaluates the operational decisions of the ACL, learns from them, and derives long-term adjustments at the policy, governance, and architecture levels. Together, both layers operationalise the resilience principle of adaptation and combine short-term responsiveness with long-term learning and development capabilities. The paper describes various implementation variants of both levels - from rule-based and KPI-driven approaches to AI-supported and meta-learning mechanisms - and shows how these can be combined depending on system complexity, data availability and degree of regulation. The proposed architecture model no longer understands resilience as a static system property, but as a continuous, data-driven process of mutual coordination and systemic learning. This creates a methodological basis for the next generation of adaptive and resilient CPSoS.