Product process coupling remains a major limitation in mechatronic system development, as nearly 80% of lifecycle cost and performance is determined during early design while manufacturing planning is still performed sequentially. Although computer-aided design (CAD) models provide rich geometric and tolerance data, and Model-Based Systems Engineering (MBSE) with SysML offers a formal system representation, these resources are rarely integrated into a unified computational workflow for intelligent process planning. This paper introduces a CAD-to-process framework that combines geometric reasoning, MBSE/SysML modeling, and reinforcement learning (RL) to automatically generate machining strategies. Functional surfaces are extracted from the CAD model and structured into SysML artifacts to encode geometric, topological, and tolerance constraints. Using the Technologically and Topologically Related Surfaces (TTRS) framework, these relations are translated into graph and matrix representations, forming the basis of a Markov Decision Process (MDP). A reinforcement learning agent then learns optimized machining sequences while preserving tolerance-chain consistency. The approach is validated on a geared-motor shaft case study, achieving 40% fewer setups, 22% reduction in redundant operations, and 31% faster convergence toward feasible process plans compared to expert-generated sequences. By reducing manual production-engineering effort, the proposed framework strengthens the integration of design and manufacturing intelligence and supports earlier, more competitive decision-making in complex mechatronic product development.
