Month: January 2013

Download PDF

This paper describes a wave based case pick operation application where a Mandate® based SOFT™ system is used to dynamically balance work between workers achieving the highest overall productivity and minimizing wave durations.

Each SKU in the facility belongs to one of twelve SKU family groups. The product is shipped to customers on pallets with most orders requiring several pallets. Family groups determine the stack sequence for the pallets, where product for higher group numbers may be stacked on lower group numbered product but not the converse.

The distribution center is divided into multiple picking zones to reduce pick congestion. SKUs from each family group are divided among the picking zones. The division keeps a single SKU in only one zone but tries to balance work in all the zones by normalizing the total velocity of all the SKUs in each picking zone. Within a particular zone, the SKUs are grouped by family and the families are more or less arranged sequentially. Multiple pickers can work in a picking zone simultaneously. Pickers receive picking instructions from RF hand held terminals.

In this particular application, product is picked to belt and delivered to a shoe sorter that diverts the cases to several “SKU family group” queues. The arriving product is dynamically assigned to the queues to facilitate proper release sequencing. Product released from the queues is delivered to another shoe sorter where it is directed to 40 manual palletizing stations.

The challenge is to complete the work from each pick zone for all products for a SKU family at essentially the same time without having workers go idle—waiting for others to complete work (both within the same zone and in the other zones).

Initially, this particular application used relative historical “worker productivity” to attempt the balance, where workers with higher productivity were partnered with those with lower productivity and then balanced against the product demand. Although this approach improved the balance, the operation still had long periods of delay where workers were idle awaiting other workers to complete.

The application was retrofitted with SOFT™ modules for labor balancing the pickers including their re-assignment across all pick zones.

While balancing labor in a system like this, there are several important factors to consider:

• Even with normalization of family group SKUs across pick zones; the random nature of the daily distribution of workload will substantially alter labor requirements in picking zones through the day.
• The number of pickers varies as workers take breaks, both scheduled and non-scheduled, and are given other assignments etc.
• Peak efficiency is accomplished only when all workers are continuously productive.
• The movement of workers between zones is inherently non-productive.
• Historical worker productivity, while significant, does not account for “bad hair” days. History was yesterday—today is reality.

Real time dynamic work balancing takes these factors into account and continuously optimizes the current resources to best meet the current requirements. With dynamic workload balancing, each time a worker completes a transaction, the SOFT™ module calculates the next action they are to take. The possibilities of the next action to take are:

• Pick the nearest item for the same SKU family in the current zone
• Pick an item for the next SKU family in the current zone
• Move to a new zone and then re-evaluate what action to take
• Wait—no new item can be picked at this time

For this particular application, the SOFT™ modules for labor balancing use the following user configurable parameters:

• SKU family delta completion time—the maximum completion time difference between zones for an SKU family. The completion time for SKU family in a zone is predicted by dividing the remaining zone-pick cases by the aggregate pick rate for all pickers currently in the pick zone. As the particular application requires, the relative work rate for a worker may be either a default value or their historical work rate.
• Workload look-ahead time—this determines how deep the pending work queue is examined for making workload balancing decisions. Pending work is organized into a prioritized queue. The queue may be literally any size. For the purpose of workload balancing it is not necessary to factor in work that is not going to be completed for a long while
• Zone change penalty—this determines how much non-productive time is required to move a worker to another zone. Depending upon the particular application this could be a singe value for any zone change or a movement velocity and an inter-zone distance table.
• Minimum zone work time—once a worker has moved to a new zone they will not be re-moved until they have worked this period of time.

These parameters coupled with the current work and the current work resources (workers) are used to control the dispatch of work. Effectively every time a worker completes a transaction, the SOFT™ module evaluates the current delta of completion times for the system. If the current delta is larger than the maximum allowed, the SOFT™ module evaluates the effect of moving the worker to each of the other picking zones. The “zone change penalty” is used in this evaluation. If moving the worker to another zone would yield a better overall completion time the worker is directed to move to the best zone, otherwise the worker is directed to complete a pick in the current zone.

As product “flows” from the pick zones it is somewhat in SKU family sequence. Due to the conveyance paths, newer SKU family goods may be placed on the conveyor before the last of the older SKU family goods have arrived (or even picked), however, the old SKU family goods are diminishing while the flow of new SKU family goods is increasing. In this application the conveyor queuing system is also dynamically controlled. As goods for a new SKU family arrive, they are diverted to a selected queue and held there until it is time for their release. If the queue fills before releasing, another queue is selected. Once a newer SKU family starts releasing, any older “straggler” SKU family goods are assigned to the queue that is releasing. The dynamic queue control allows multiple SKU families to be simultaneously queued while allowing large SKU families to occupy space as necessary in multiple queues.

This labor-balancing algorithm increased worker productivity by over 15% in applications where the previous balancing used measured historical productivity rates to statically determine work organization. Overall improvements resulted in reduced idle time for the workers, more sequential flow of SKU family groups to palletization and increased processing capacity for the facility.

Download PDF

In retail operations, most sales occur during the weekend. Orders for replenishment goods arrive at the DC early in the week and must be delivered to the retailer in time for the next weekend activity. The DC has Monday and Tuesday to process and ship the orders. Scheduling of order processing within these two days is dictated by the distances that the goods must travel to reach the retailers: orders processed first are those shipped to the farthest locations, while the orders processed last are those shipped to local customers.

As the market shifts to more frequent deliveries of smaller orders with shorter lead times, two implications critical to DC operations emerge: (1) pre-booked orders decline as at-once orders rise, and (2) orders processed as full-case quantities decrease, or even disappear, as orders requiring piece picking from broken cartons become the norm:

• Pre-booked orders with required lead times of several days or even weeks allowed DC’s to control the order processing schedule, making the operation of the DC more efficient. At-once orders must be processed and shipped as they are received, preventing any advanced scheduling that could optimize the DC’s operation.
• The differences between full-case processing and piece-picking operations are well known; while full-case quantity processing allows the handling of many units with little effort, piece picking is very labor intensive, requiring many resources for handling small quantities.

For DC’s with manual operations, a shift to piece-picking-intensive transactions represents higher labor requirements. As labor costs reach unacceptable levels or as workers begin to literally run into each other because the infrastructure cannot support the additional personnel required to satisfy the customer orders, DC managers are forced to consider other modes of operation.

By adding automatic sortation for order consolidation to a manual system, a DC can process orders in waves, or batches, fulfilling several orders at the same time. In doing so, the units to pick per SKU per transaction increase, allowing the DC to restore some of the full-case quantity processing lost with the new market requirements. However, when the orders being fulfilled are composed of small quantities per SKU per order, retrieving product to feed a sorter still can be very demanding. DC’s may purchase a sorter only to discover that they lack sufficient retrieval capacity to feed it.

In my article “Designing Distribution Centers: Shifting to an Automated System,” I presented equations to calculate throughput requirements for automated systems with sorters. One of the conclusions from that article, namely that the segregation of the storage according to carton condition (full-case or residual) optimizes the feeding of the sorter, is the concept underlying the QARS model for DC design.

When we analyze operational data of a DC collected over an extended period of time—say, six months or a year—we typically find that a small portion of the DC’s available SKUs, known as fast movers, account for the majority of the DC’s activity. Similarly, when we analyze the same operational data at a wave level, we see that in each wave a small portion of the total SKUs required to fill the wave’s orders accounted for a large portion of the wave’s total units. Further analysis also indicates that:

• The fast-moving SKUs for a particular wave do not necessarily match the fast-moving SKUs in the DC’s total inventory.
• The slow-moving SKUs for a particular wave do not necessarily match the slow-moving SKUs in the DC’s total inventory.
• The fastest moving SKUs in a wave change from wave to wave.

Traditional designs segregate inventory based on the activity level of SKUs, fast-movers or slow-movers. Using this concept, fast-moving SKUs are placed in storage devices where they are easily accessible, facilitating the majority of the transactions. However, this approach presents some complications. To illustrate, let us assume a DC operation where 20% of the SKUs account for 80% of the activity:

• If the fast-moving SKUs represent 80% of the DC’s activity, they will also represent 80% of the inventory on hand. If we place all the fast-moving SKU inventory in the fast-moving storage device, we would require storage capacity for 80% of our stock. We may as well have all the inventory in the fast-moving storage device.
• If we split the fast-moving inventory between a small active area equipped with the fast-moving device and a large reserve area, we solve the problem of a large storage capacity requirement for the fast-moving storage device. However, we must now double handle all fast-movers (80% of our activity) because of the replenishment requirements of the active zone from the reserve zone.
• A DC’s fast-moving SKUs are not always the same SKUs. The activity level classification for each SKU must be frequently re-evaluated to see if the SKU still belongs in its current storage area (fast-moving or slow-moving). If SKU classification changes, all inventory with that SKU must be moved to the other storage area.

The QARS model, described in the following section, resolves the complications associated with activity-based segregation of inventory.

The QARS Model

The QARS model exploits an operational strategy that segregates inventory based on the carton quantity, full-case cartons or residual cartons (previously open cartons with less units than the original full-case quantity), regardless of whether the carton contains fast- or slow-moving SKUs. It does not require any planning or additional handling for the inventory in residual storage.

In piece-picking operations, transactions related to residual cartons account for a large portion of the DC activity and an even higher portion of the required resources. In a QARS-based DC design, the handling of the residual cartons is automated, which greatly simplifies a piece-picking operation.

Let us consider a QARS-based DC with:

• A sortation system to which both full and residual cartons are conveyed, and where operators at sorter induction stations pick the units required for the wave in progress
• A bulk storage area made of conventional rack with sufficient storage capacity for all full-case cartons in the facility
• A residual storage area consisting of a high-throughput AS/RS for residual carton retrieval and putaway, with sufficient storage capacity for at least one residual carton per SKU

The system is operated as follows:

• If the requested quantity for a SKU is less than the quantity available from the carton in the residual storage, the units are inducted from the residual carton and the carton is returned to the residual storage.
• If the requested quantity for a SKU is equal to the quantity available from the carton in the residual storage, all units are inducted from that carton. The empty carton is removed from circulation.
• If the requested quantity for a SKU is larger than the quantity available from the carton in the residual storage, all units from the residual carton are inducted and the empty carton is removed from circulation. Additional cartons are conveyed from the bulk storage area and units from full-cases are inducted until the requested quantity is reached. Empty cartons are removed from circulation. If there are residual units in the last carton, this carton is sent to residual storage.

Example 1

Full-case carton quantity = 25
Units in residual storage carton = 10
Requested quantity = 7

Transactions

• The residual carton with 10 units is transferred to the sorter induction workstations, and 7 units from this carton are transferred to the sorter.
• The residual carton, which now contains 3 units, is returned to residual storage.

Example 2

Full-case carton quantity = 25
Units in residual storage carton = 3
Requested quantity = 38

Transactions

• The current residual carton with 3 units is transferred to the sorter induction workstations and completely depleted, that is, 3 units from this carton are transferred to the sorter.
• Two full-case cartons from bulk storage are transferred to the sorter induction workstations. All 25 units from the first carton are transferred to the sorter plus 10 units from the second carton.
• The second carton, now the new residual carton with 15 units, is sent to residual storage.

The QARS model is flexible enough to support specific DC requirements—for instance, if a DC is required to never fully deplete SKUs in the residual storage area, the design can be altered to allow for the temporary presence of more than one carton per SKU in the residual storage.
A QARS Case Study
To help illustrate the application of QARS in a real-world situation, consider the following case study:

A DC receives and stocks full-case quantities of product (pure-SKU receiving cartons) and consolidates smaller quantities of multiple SKUs for customer delivery (mixed-SKU shipping cartons). The DC does not have control over its retailers’ orders, so it cannot arbitrarily ship the residual units from open cartons to empty them. All shipped units must be the SKUs and quantities specified by the retailers. For design purposes, all orders are assumed to be at-once orders. The operation consists of:

• 8,000,000 shipped units per year
• 20 units per receiving carton (average)
• 1,000,000 units required storage capacity
• 10,000 available SKUs in storage
• 80,000 units shipped per day (design and peak day)
• 7 operating hours per day

When designing DCs based on historical data, it is common practice to select a very busy actual day of the DC’s recent past operation and to use the activity of that day as the design basis. I find it rather risky to base an investment of several million dollars on such limited data. Even if the shipped volume on the selected day is equal to the design value, there is no guarantee that the shipping profiles (number of active SKUs, for instance) of the selected day should be used for the design. A safer practice is to analyze the shipping profiles over a longer period of time and, using statistical tools and simulation modeling, create multiple “design days” based on those profiles, expected changes, and the design shipping volume. Later, if desired, the design model can be validated with the data from any historical day.

To define the design order profile, historical shipping data from the DC’s nine consecutive busiest weeks was used. Important statistics from this period include:

• 1,900,000 units shipped
• 30.1 units per order (average)
• 10.1 SKUs per order (average)
• 2.98 units per SKU per order (average)
• 2,000 SKUs (fast movers) account for 80% of the DC’s total SKU activity

In my previous article, I explained the importance, as a design parameter, of the units to process per wave (and the corresponding expected SKUs per wave): A large number of units per wave implies a larger sorter that requires less AS/RS throughput capacity to feed it, while a small number of units per wave implies a smaller sorter that requires more AS/RS throughput capacity.

Through simulation, and using the historical shipping profiles, the expected number of active SKUs per wave, as a function of the wave size (in units), was calculated for this particular case study. The following chart presents the results of these calculations:

An explanation of how to determine the optimal operating conditions for such a system can be found in my previous article. Let us assume that the optimum operating conditions for this DC require seven one-hour waves per day (an average of 11,426 units per wave).
If the DC operates under design, or peak, conditions an average of 20 days per year, a simulation of 600 days under peak conditions would be equivalent to checking the DC’s worst conditions for 30 years. Under these operating conditions, the simulation model yields the following throughput requirements:

The corresponding storage, based on design requirements, is:

(*) Full-case cartons have all their original units, residual cartons have half their original units.

(**) Storage utilization is storage locations with cartons divided by total storage locations.

Today’s emerging technologies, which focus on throughput rather than storage, provide devices capable of up to 750 dual transactions per hour (one dual transaction is equivalent to one retrieved carton and one putaway carton) at reasonable prices.

For residual cartons in this case study, we can use AS/RS carousels, each with a throughput capacity of 600 dual transactions per hour and 2,630 carton locations for storage. Four of these AS/RS carousels provide all the throughput and storage capacity required for residual cartons.

The simulation indicated some instances where the required throughput from the AS/RS carousels exceeded 2,400 dual transactions per hour. Further, some imbalance between the requirements for each AS/RS carousel during the waves is to be expected, as well as some inefficient utilization of the AS/RS carousels during transition between waves. These conditions extend the duration of some waves beyond the allowed 60 minutes. Some options to deal with these situations are: (1) allowing some overtime at the end of the day, (2) a buffer (like conveyor queues) between the AS/RS carousels and the sorter, and (3) more throughput capacity (five AS/RS carousels instead of four).

The number of SKUs that require manual handling during a wave, which can be related to the number of locations to visit in a manual system, is reduced from 2,059 to 517. More importantly, the eliminated SKUs are those associated with dual transactions (one retrieval and one putaway); the remaining SKUs are associated with only full-case retrievals.

Despite having an 87% piece-picking operation, all manual product retrieval is full-case (piece counting still is performed manually at sorter induction). Through the automation of only 15% of the DC’s total storage, 89% of the required handling is automated, without any double handling of product.

The DC can accept waves, or batches of orders to ship (up to the capacity of the sorter), every hour. Two and a half hours after receiving a batch, regardless of SKU mix, all the orders can be in trailers, packed and loaded, ready to leave. The QARS-based system has no dependencies on any kind of wave balancing.
Summary
A QARS-based system design allows a DC to respond immediately to any set of orders with very short notice. Processing of orders can be scheduled based on customer needs rather than the DC’s available resources.

Transactions between the residual storage area and the sorter must be dual transactions at the time of order processing, as this is the kind of transaction supported by AS/RSs. In contrast, transactions from bulk storage racks may be single transactions, as the complementary transaction—that is, putaway from receiving—can occur at non-peak periods. This condition allows the DC to manage and schedule labor resources in the most efficient manner.

Product in the AS/RS carousels need not be continually re-evaluated; residual cartons belong there and full-case cartons do not. All cartons are touched only three times: (1) at receiving, (2) when they are put away in the bulk storage rack, and (3) when they are taken to the sorter. Replenishment of the AS/RS carousels is automatic, simple, and does not require any planning or additional handling of the product.

Each DC is unique, with its own specific needs and requirements. It would be naïve to presume that there is a universal solution for all the possible situations. Nevertheless, the QARS model provides a good design base for those operations in which increasing piece-picking requirements and decreasing required processing times are forcing the DC managers to contemplate alternative modes of operation.

Download PDF

Two of the main features of SOFT-based smart carts are the very efficient way in which they dynamically organize work and their flexibility to “adapt” to random operation conditions that cannot be planned in advance.

These two features make these carts excellent devices to efficiently support one of the toughest functions of distribution centers: the re-stocking of small quantities of various items or products as encountered in returns processing operations.

This paper presents two separate application examples of returns processing using smart carts. In the first application, the restocking operation product goes into a single storage zone. In the other application, there are several storage zones where re-stocked product is to be placed. The second application uses some of the standard features of a SOFT system to pre-sort product into zones using a smart cart, and then use the carts to re-stock items in their individual zones.

It is assumed that the items to be put-away are bar coded. In applications where the items are not bar coded, workers may use visual identification/validation instead of scanning the items. In this case SOFT systems offer features to create SKU barcode labels. If this feature is used, savings in time and errors for the actual put-away operation normally offsets the additional time used in the creation and attachment of the labels. Workers inspecting the condition of the returns can label them as they are identified.

Both applications use a smart cart with 20 totes; however, as with all SOFT applications, the number storage locations (i.e. totes) can be easily subdivided into sub-cells and multiple SKUs can be contained in a single storage location if so desired. The smart cart has lights on both sides of the cart to identify each tote.

Single Storage Zone Application

After selecting the appropriate SOFT menu item for put-away, the smart cart worker individually scans items to be put-away. The worker then normally scans the item into a storage location or a cell. Since SOFT is continuously “optimizing” work, if an item is scanned that matches an item already on the cart, SOFT would like to “consolidate” the items in the same cell. In this case, when a matching item is scanned, SOFT immediately produces an audible alert sound on the cart and lights the tote location of the existing item while indicating the cell on the cart display. The worker would then normally place (and scan) the matching item with its companions, however the item may be scanned into any cell as desired. Totes may be sub-divided (and un-divided) “on the fly” as desired. This is done through the use of storage location creation and deletion menu items. SOFT systems support multiple SKUs in a storage location; and depending on the desired operation parameters, a warning may be issued to the user to indicate a mixing of product. Together these features are used as required to fill the cart to capacity. The greater the fill capacity of the cart, the more efficient the put-away operation will become. When the cart is continuously reloaded as the returned items are put away, the walking is reduced almost by half!

The storage cells on smart carts are designated in SOFT as temporary storage locations. Thus the adaptive nature of SOFT has the inherent objective of placing the items in permanent locations. When the put-away menu item is selected on a cart and any permanent storage location is scanned, SOFT dynamically optimizes the list of put-away items based on the current location of the cart and direct the worker to the nearest location to be re-stocked for an item on the cart. The cart location containing the item is illuminated as well as the location identified on the smart cart display. The worker is informed if the location (tote) for the item to be put-away contains multiple SKUs, improving efficiency. The worker retrieves the specified item and then scans it and then scans the specified put-away storage location. Once completed, SOFT selects the next nearest item on the cart for put-away and the operation is normally continued until the cart is emptied. Like SOFT picking operations, the current location of the cart may be modified at any time by scanning any permanent storage location. This will re-optimize the remaining work on the cart based on the scanned location. While the normal put-away procedure may be to fill the cart and then completely empty the cart, the adaptive nature of SOFT allows new items to be added to the cart dynamically or “on the fly” as local operational procedures dictate.

Multi Storage Zone Application

This example application is similar the prior example except it is a two step operation, where a smart cart is used to “pre-sort” items into zones for later zone put-away. Although a two-step operation, the efficiency of the put-away process is improved by reducing the walk time, since for the actual put-away process the cart is normally operated in only a single zone. In this application, the presort uses individual smart cart locations as “virtual zone staging locations” where zone totes are filled. The filling of the totes uses the same procedure as in the example above except the “combining” of like items is on a zone basis rather than a SKU basis. As a tote for a particular zone is filled, that tote is removed and placed in an area with the previously filled totes for that zone. A new empty tote is then scanned into the smart cart location and that new tote is assigned to the zone. New items scanned are directed into the cart locations corresponding to the put-away zone.

When zone put-away is to commence, totes filled for a particular zone are loaded and scanned into a smart cart for put-away. The put-away process proceeds as in the first example. Totes for different zones may be loaded onto a cart as local procedures specify. The smart cart will automatically and dynamically optimize the put-away operation to most efficiently empty the cart.

Conclusion

Smart cart technology provides a low capital investment option to increase picking productivity. SOFT based smart cart benefits are not limited only to picking; because of its flexibility and how they adapt to different operation requirements they are also excellent devices to process returns. This paper presented a few options to handle returns with SOFT based smart carts. How can you take advantage of these carts’ flexibility to process your returns?