A reinforcement learning approach to production planning in the fabrication/fulfillment manufacturing process

We have used reinforcement learning together with Monte Carlo simulation to solve a multiperiod production planning problem in a two-stage hybrid manufacturing process (a combination of build-to-plan with build-to-order) with a capacity constraint. Our model minimizes inventory and penalty costs while considering real-world complexities such as different component types sharing the same manufacturing capacity, multiend-products sharing common components, multiechelon bill-of-material (BOM), random lead times, etc. To efficiently search in the huge solution space, we designed a two-phase learning scheme where "good" capacity usage ratios are first found for different decision epochs, based on which a detailed production schedule is further unproved through learning to minimize costs. We illustrate our approach through an example and conclude discussion of future research directions.