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

The primary performance measure used in the study is the average travel time of the AS/R machine for storing and retrieving a unit load. Utilization of the AS/R machine, shelf utilization, and service level of the system are also considered as secondary measures.

4.2. Results of the experiment

In the experiment, the four policies are compared for each combination of the remaining four factors. The average and the variance of observations are calculated and analyzed over five replications. The complete set of results for all combinations can be found in Bozkurt [21]. These results are summarized below.

When the order quantity and the ABC factor are fixed but the number of products is varied, the following observations can be made which are common to 10, 20, and 30 product type cases.

* At full capacity, the best storage assignment policy in terms of the average AS/R machine travel time is DCL. DCL is followed by TCL and DED which are very close together but significantly worse than DCL. RAN yields the longest travel times. The situation is better amplified with ten product types.

* Shrinking the capacity always results in a significant reduction in the travel time. This is expected because travel distances get shorter as the AS/RS gets smaller.

* Capacity shrinkage also results in convergence of the three shared storage policies. The policies become indistinguishable at 80% of full capacity. The convergence of TCL and RAN is faster than that of DCL.

* The AS/R machine utilization decreases along with the travel time. Shelf utilization, which does not vary significantly by the policy, increases as the capacity is shrunk.

When the number of products and the ABC factor are fixed, and the order quantity is set as 20, 30 or 40, the trends observed are identical to those given above. When only the ABC factor is changed, keeping the number of products and the order quantity constant, the results are again similar. In summary, DCL proves to be superior to other storage assignment policies under the present slot allocation scheme to unit loads, i.e., assigning one common slot to the unit loads having the same expected DOS.

5. Experimental results with the continuous review (Q,r) inventory policy

We have applied the continuous review (Q,r) model to determine the optimal values of order quantity and reorder point for each item under Poisson demand. We have used Hadley and Whitin's [22] algorithm for multiple items, assuming that there are no interactions among items. Our simulation model, which is adapted from Bozkurt [21], is run using these order quantities and reorder points for replenishment of products in the AS/RS. A detailed description of the revised model can be found in Kulturel [23], and the main features are summarized in the Appendix.

From this point on, we focus on the two shared storage assignment policies, TCL and DCL. DED and RAN are eliminated from further consideration because they prove to be inferior according to the results of Section 4.

As in Section 4, products and racks are divided into three classes according to ABC classification, and partitioning of the rack space is with proportions 20/40/40. However, in implementing DCL, a different technique is used for dividing unit loads into three classes. Once the expected DOS of each unit load is calculated, these values are sorted in nondecreasing order. Among the unit loads, the first 20% having the shortest DOS values are assigned to class A, the next 40% to class B, and the last 40% to class C. If some unit loads immediately following the 20% quantile have the same DOS as the previous one, these are also included in class A.