Predictive Control for Skill-Constrained Manufacturing Supply Chains

A new model predictive controller has been developed for manufacturing and supply chain systems that explicitly accounts for workforce skills. This controller operates in a closed-loop fashion, making decisions on production, inventory, backlog, and training at each shift by solving a finite-horizon mixed-integer program. The model incorporates realistic constraints such as skill certification requirements, the decay of certifications over time, and the fact that training consumes the same valuable worker hours needed for production. It uses an interpretable terminal value to price certified-capacity gaps at the horizon boundary. Evaluations on SkillChain-Gym scenarios, including various shocks and forecast modes, show that no single policy class consistently outperforms others. Predictive control is most effective when skill or labor bottlenecks are forecastable, allowing sufficient time for training. However, lean static insurance plans prove more robust against surprise shocks and near capacity limits.