A Common Experimentation Platform for Engineer-to-Order Production Planning: Optimization Patterns as a Stable Core, Solver and Data Interfaces as Boundary Resources

Production planning in Engineer-to-Order (ETO) companies is a multi-project scheduling problem with long lead times, overlapping engineering and production phases and scarce shared resources. New planning methods are hard to evaluate in this setting: company data are confidential and costly to prepare, so each study builds its own model, data and solver pipeline, and results are rarely comparable. This paper applies platform architecture concepts to this problem and presents a common experimentation platform developed within the publicly funded ETOSS project. A library of optimization patterns and a canonical ETO data schema form the stable core. The solver and data interfaces are boundary resources, against which solvers, learning-based planners and industrial domains are connected as complements; a calibrated instance generator replaces confidential data with shareable synthetic benchmarks. A proof of concept shows that two constraint programming solvers were connected to the same core with changes confined to their bridges, that a second domain was added as a phase template, and that every discrepancy found lay at the interfaces. From these discrepancies the paper derives a first specification of the solver interface, and it closes with a research agenda for the remaining project activities.