Robotic assembly in flexible manufacturing faces challenges in adapting to dynamic environments and handling system failures. These challenges necessitate swift identification and execution of recovery processes without impeding ongoing operations. Furthermore, the capability limitations of existing robots require adaptive solutions that integrate additional robots and implement compensation actions. Model-driven development offers significant advantages for robotic assembly systems, particularly in maintaining consistency, facilitating recovery process identification, and seamlessly integrating new plug-in robots, yet challenges remain in fully leveraging these benefits in dynamic manufacturing environments. Addressing these challenges, this paper presents a novel constraint-based model using Reference Attribute Grammars (RAGs) to extend an existing knowledge-based approach, enabling self-healing capabilities in robotic assembly processes. Our model integrates with a MAPE-K feedback loop, facilitating real-time adaptation to environmental changes and system failures. By combining constraint checking and assembly sequence planning, our approach allows for efficient fault detection, diagnosis, and recovery. A key innovation of our work is the implementation of a modular, extensible constraint model that defines constraints for robots, components, and processes. This design enables incremental reevaluation of constraints and assembly processes, supporting continuous monitoring, analysis, and replanning. We demonstrate the efficacy of our self-healing capabilities through two scenarios involving multiple robots with diverse capabilities collaborating on product assembly. These scenarios showcase the system’s ability to dynamically reorganize plug-in robots and reallocate tasks and resources in response to failures.