Product Cost Estimation for Manufacturing Platforms

Manufacturing-as-a-service (MaaS) platforms require fast and accurate manufacturing cost estimation to support quotation and resource allocation across distributed networks. However, existing machine learning approaches are often constrained by limited data availability, proprietary information, and insufficient interpretability. This work presents an engineering-based framework for generating high-fidelity synthetic data to develop scalable and interpretable cost estimation models, culminating in a SHAP-interpreted XGBoost model. A parameterized prismatic workpiece was sampled using a Hammersley sequence design of experiments, producing 1355 geometrically feasible parts with representative drilling and pocket milling features. For each configuration, cutting tools, machining parameters, and CNC toolpaths were automatically generated and optimized through an integrated SolidWorks, MATLAB, and VERICUT workflow. Over 70,000 physics-based simulations were performed to obtain reference values for machining time, tool usage, and manufacturing cost. The resulting dataset integrates geometric descriptors and optimized process parameters, enabling the training and evaluation of machine learning models. The proposed methodology demonstrates that engineering-driven synthetic data provides a physically consistent foundation for fast and interpretable AI-enabled cost estimation. While applied to a specific family of three-axis milled parts, this approach serves as a proof-of-concept for predictive micro-services required in next-generation platform-based manufacturing systems.