Predictive vehicle dispatching algorihtm for overhead hoist transport systems in semiconductor fabs

Abstract

In this thesis, two novel predictive dispatching rules for the Overhead Hoist Transport (OHT): Basic Predictive Dispatching (BPD), Certainty Weighted Predictive Dispatching (CWPD) are proposed. Unlike conventional OHT dispatching rules, the proposed rules additionally consider two kinds of information: the expected arrival time of near-future transfer requests and the time required for delivery vehicles to be idle. BPD is first proposed for the environment where the prediction on the arrival time of transfer requests has no error. To improve the performance of BPD in the environment where the prediction is not perfectly accurate, the enhanced version of BPD called CWPD is proposed. Both BPD and CWPD formulate the OHT dispatching problem as a linear assignment problem while they employ different matching cost functions. The effectiveness of our proposed rules is tested in a virtual semiconductor fab. In the environment without prediction errors, experiment results illustrate that BPD outperforms Shortest Travel Distance First (STDF) rule and Hungarian Algorithm Based Reassignment (HABOR) rule by a wide margin, which verifies the value of the two kinds of information for OHT dispatching. In the environment with prediction errors, the superiority of CWPD over BPD is confirmed, which indicates the higher practicality of CWPD in the real world.