Experimental investigation of shared storage assignment policies in automated storage/retrieval systems

IIE Transactions, August, 1999 by Sadan Kulturel, Nur E. Ozdemirel, Canan Sepil, Zafer Bozkurt

We conclude that TCL consistently outperforms DCL, yielding shorter AS/R machine travel times compared to DCL. However, the difference between the two policies is not very large in many cases. In particular, as the number of product types gets larger, AS/RS capacity is shrunk, and the system becomes more balanced, the difference between the two policies becomes negligible.

We can summarize the effects of our experimental factors as follows.

* In order to reduce both the AS/R machine travel time and the balance of the system, we must use the smallest possible replenishment order quantities. This, however, results in slightly higher backorder levels.

* We also need to use the smallest possible AS/RS capacity to keep the travel time short. Shrinking the capacity down to 80% of the full capacity does not cause a significant change in service level, balance, or number of backorders. This also results in convergence of the two storage assignment policies.

* Neither the travel time nor the balance of the system seems to be significantly affected by the change in the backorder cost.

* Finally, both the travel time and the balance are reduced if the ABC factor in a given system is at a relatively high level. This causes the number of backorders to increase.

The following conclusions can be drawn regarding the sensitivity analysis.

* The difference between the two storage assignment policies becomes negligible as the number of product types increases.

* The relative behavior of TCL and DCL remains unchanged for different settings of the replenishment lead time. Typically, travel time, balance, shelf utilization, and service level increase when lead times are zero for all items. The number of backorders, on the other hand, decreases in this case. Doubling lead times has a reverse effect on the performance measures. * Doubling the demand rate has a similar impact on performance measures as doubling the lead time. In either case, as the system becomes more balanced, the difference between the two policies diminishes.

When we consider the effects of the inventory policy and the ABC classification technique, we can conclude that using the (Q,r) policy to determine the order quantities, choosing TCL as the storage assignment policy, and partitioning the rack space into classes with proportions 10/45/45 yield the shortest average AS/R machine travel time.

Our research is restricted to a single AS/R machine operating in a single aisle AS/RS. Although this system is representative of a larger AS/RS with multiple aisles and AS/R machines, it will still be useful to extend the ideas developed in this study to a larger system. As the system gets larger, the measurements will be more on a continuous scale, reducing undesirable effects of rounding due to DOS ranges or rack partitioning in multiples of five. Also, this would bring about new issues such as routing and collision avoidance of multiple vehicles, and an even distribution of traffic throughout the AS/RS.

An important extension of this research would be the application of dynamic storage assignment policies. A dynamic storage assignment policy assigns unit loads to the next available class whenever the designated class is full, thereby adjusting the borders of rack partitions. This approach provides better utilization of space and higher service level. Moreover, this may provide an opportunity for shrinking the capacity more than we could under the currently used policies.