Receding Time Horizon Self-Tracking and Assessment for Autonomous Manufacturing

Next-generation production and manufacturing plants create individualized products by automatically deriving production schedules from design specifications. However, because planning and scheduling are computationally hard, they must typically be done off-line using a simplified system model, and are consequently unaware of on-line observations and potential component faults. This leads to a self-assessment problem we call plan assessment: Given behavior models and current observations of the plant's (possibly faulty) behavior, how likely is it that partially executed manufacturing plans will still succeed? A previously developed self-assessment capability generates k most probable system behaviors as solutions of a constraint optimization problem. However, it only allows single products and doesn't scale well with the schedule length. In this work, we extend the capability towards multiple products and generate k most probable system behaviors only within a small, fixed time window, which is then moved to cover schedules longer than the time window. Experiments show the feasibility of this approximative approach.