Scientific discovery increasingly demands methods that can cope with complex systems and high-dimensional experimental spaces. Traditional iterative approaches—rooted in hypothesis, experimentation, and modeling—struggle to scale under such conditions. This paper introduces a conceptual framework that combines two digital twins to accelerate and systematize the scientific discovery process. The first, a Phenomenon Digital Twin, simulates the physical system under investigation. The second, a Scientific Discovery Digital Twin, uses Generative Flow Networks (GFlowNets) to intelligently explore the experimental design space and prioritize informative experiments. This dual-DT architecture enables iterative refinement of models while optimizing data collection under budget constraints. The framework is demonstrated through an illustrative case study on plasma-enhanced deposition, a materials science domain characterized by poorly understood phenomena and large configuration spaces. While the proposed approach is still in its conceptual stage, it outlines a pathway toward adaptive, AI-assisted scientific exploration applicable across disciplines.