Estimating job flow times by using an agent-based approach

We consider the problem of estimating flow times of jobs that arrive dynamically in a manufacturing system. A job's flow time refers to the time between the job's arrival and completion. Most existing methods use some predefined equations for such estimation, and most of the equations are designed for single machine manufacturing systems. To better estimate the flow time of a job in a more complex system in which there are multiple machines and multiple workstations, we propose a distributed learning approach that divides the manufacturing system into multiple small parts and collects real-time local information in each part to predict the waiting time for a job. We evaluate the proposed approach by comparing it with existing methods using a variety of problem instances. The results show that the proposed approach outperforms existing methods and accordingly might improve the level of customer service when being used for due date promising.