Architectural Patterns for Integrating AI Technology into Safety-Critical Systems

Artificial Intelligence (AI) is widely acknowledged as one of the most disruptive technologies driving the digital transformation of industries, enterprises, and societies in the 21st century. Advances in computing speed, algorithmic improvements, and access to a vast amount of data contributed to the adaption of AI in many different domains. Due to the outstanding performance, AI technology is increasingly integrated into safety-critical applications. However, the established safety engineering processes and practices have been only successfully applied in conventional model-based system development and no commonly agreed approaches for integrating AI technology are available yet. This work presents two architectural patterns that can support designers and engineers in the conception of safety-critical AI-enhanced cyber-physical system (CPS) applications. The first pattern addresses the problem of integrating AI capabilities into safety-critical functions. The second pattern deals with architectural approaches to integrate AI technologies for monitoring and learning system-specific behavior at runtime.