Motorola Boosts Distribution Centre Productivity

Motorola increases productivity of distribution centre by 150% in only six months using WITNESS simulation software from the Lanner Group.

Background

The Motorola Radio Products Group (RPG) provides high-quality, mobile and portable two-way radio products to end users through a network of local distributors and dealers. Typical users of RPG's products include construction firms, delivery and service companies, police and security organisations, transport and taxi companies, consumer and light industrial users and small businesses. Also, consumers are increasingly recognising the benefits of two-way radios for personal or recreational applications. Two-way radios provide instant connectivity—whether it is “one-to-one” or “one-to-many”—for both on-site applications and wide-area communications.

Motorola opened its Suwanee, Georgia distribution centre in May of 1997 to support consumer and light industrial users of Motorola’s land mobile subscriber group. Initially, the operation was designed to pick, pack and ship radio and accessory orders. As the product line increased, operations were added to configure orders to specific customer packaging and accessory requirements. These operations were called postponement. The new operations combined with higher sales dramatically increased the workload of the centre. Since it was a new facility, there was no historical data that would allow management to benchmark operating efficiency.

Motorola technical staff decided to use discrete event simulation to model the operation of the facility and substantially improve Suwanee distribution centre operations. Motorola engineers selected WITNESS as the simulation engine for several reasons. First, WITNESS provides a simpler and more flexible user interface than the other packages we examined. Since the interface is graphical, the user can model a distribution process without writing any code. Second, WITNESS provides a powerful optimisation module that provides the ability to rapidly perform “what-if” experiments in order to explore alternative business scenarios. Model parameters that may be optimised include processing times, quantity, staff numbers, queue sizes, equipment speeds, etc.

Wave release

The “wave release” is the central forcing function within the model. A wave is a collection of orders released to the floor that optimises the picking, packing and consolidation requirements with inventory availability. In practice, waves are manually generated as orders are placed on the distribution centre. In the model, waves are generated one at a time and each wave is not started until the first has completed the total picking function. All material is available for picking. Each wave has associated with it the quantity of picks by type required to complete it. Distributions were developed using the data collected by the automated operating system.

The model started with trucks arriving at the receiving dock and ended with loading trucks at the shipping dock. Four types of direct labour classifications were identified: receiving, shipping, picking and postponement labour. Receiving personnel unload inbound material to bulk, replenish picking stock and make some case picks. Picking personnel make all other picks including radio, accessory and remaining case picks. They may be used to consolidate orders if the picking requirements don’t use all the headcount and work is available in the consolidation area. Shipping personnel consolidate orders and load outbound trucks. Postponement personnel perform all radio and accessory postponement operations. Since all floor locations had to be restocked from bulk stock, stock shortages could occur at the picking stations or in bulk storage.

Parameter optimization

WITNESS Optimizer was used to automate the evaluation of a wide range of business scenarios without the need for user intervention. The following parameters were varied to generate the scenarios: 

1. % case picks were varied from 0.5% to 2% in increments of 0.5%
2. % accessory picks were varied from 25% to 49% in increments of 8%
3. receiving headcount was varied from 3 to 9 in increments of 3
4. floor picking headcount was varied from 11 to 20 in increments of 3
5. shipping headcount was varied from 4 to 8 in increments of 2
6. postponement headcount was varied from 7 to 13 in increments of 3.

The simulation clock was set to 9,000 minutes or 20 workdays of 7.5 hours each. Less than four days were required to execute the 2,304 computer simulations representing all possible combinations of the input variables. As each of the simulations was completed, the critical output measurements were written to a file. Output variables included: 

1. % case
2. % radios
3. % accessories
4. stock headcount
5. floor headcount
6. shipping headcount
7. postponement headcount
8. number of waves generated
9. number of postponement picks
10. number of case picks
11. number of radio picks
12. number of accessory picks
13. number of FP5 picks
14. total picks
15. calculated stock headcount
16. calculated floor headcount
17. calculated shipping headcount
18. calculated postponement headcount
19. total cartons
20. picks per floor headcount
21. cost per pick.

Response surface equation

The model was validated by comparing the resulting output to the actual distribution centre performance. When the output set was complete, the file was read into the JMP statistical discovery software package developed by SAS Institute, Inc. This package is used to find values for one or more factors that maximise or minimise the response of a nonlinear function. JMP provides the tools to fit response surface models and analyse the surface. Response surface methodology fits parabolas or paraboloids to the response and finds the optimum value on the fitted surface. In this application, JMP was used to generate the response surface equation used to predict the number of picks for accessories and radios. The equation explains 99.7% of the variation.

To use the equation developed by the parametric estimates, the number of waves was required. This parameter was not normally forecasted at the distribution centre because it is very arbitrary in content, being based on the combining orders as they are received. An estimate of the number of waves as a function of total picks or radios shipped would then cascade into the JMP equation.

Results

The two equations formed the basis of an Excel spreadsheet tool that allows management to determine the operating efficiency of the plant in seconds. The projected accuracy of this tool to forecast headcount requirements is about 95% in actual headcount. All they have to do is enter the product mix and head count values to determine the expected number of waves and picks. 

This tool was extremely useful because, with the plant in a start-up mode prior to its development, it was nearly impossible for management to determine how efficiently the plant was operating and whether changes were having a positive or negative impact because of the constantly changing workload and product mix. By giving managers instantaneous feedback on the facility's performance, the spreadsheet tool allowed them to rapidly improve distribution centre efficiency.