Helia Jalili, M. Reza Yaesubi, Cavosh Innovation, Newton, MA
Despite growing interest in AI-driven process optimization, PVD process development remains largely iterative, requiring multiple deposition runs to establish a stable process window for new coatings or target specifications. Modern coating systems continuously generate process data, yet this data is rarely used beyond traceability — raising a practical question: if the data already exists, why is it still not used predictively? This talk examines that gap through the lens of real-world industrial implementation. In practice, process records are distributed across logs, reports, and operator notes, often lacking the consistency and input–output linkage required for predictive modeling. A structured workflow for moving from raw historical process records to predictive process guidance is presented, with emphasis on the data preparation steps that must precede modeling. The discussion highlights both the opportunities and the practical limitations observed in production coating environments.