The more people contributing articles, ideas on articles or even just notices they want to have generally communicated, the more successful the newsletter will be. So please if you have something you want see in the newsletter or an opinion on how it could be improved please contact me.
Brian Tomlin
Massachusetts Institute of Technology,
Building E53-364,
50 Memorial Drive
MA 02142
Phone : 617-253-6638
Email : btomlin@mit.edu
The group had a meeting at MIT in mid-March. As the whole group could not attend, George Daddis of Kodak has some questions he would like to communicate to the group :
Subject: Questions for S/L Partners
In manufacturing, Kodak's cycle time efforts are closely linked with supply chain improvement (re-engineering efforts of product flows like Kodacolor film are good examples of this link). We also have product specific SCOT (or Supply Chain Optimization Teams) whose focus is process cycle time, and have had LFM interns working with these teams in the past.
I'd be interested in understanding how other partner companies are dealing with this general area. The questions I was going to raise for general discussion at this week's meeting were:
- Do partner companies have a formal cycle time initiative as part of their Total Quality effort?
- Is there a strong link in manufacturing between cycle time and supply chain improvement/optimization?
- Do partner companies work to a specific target that is somehow based on the "value adding steps" (theoretical time) contained in that flow? (Examples that come to mind are Phil Thomas' Entitlement or Realizable Performance, DEC/Cathy Nash's A-Delta-T, J&J's A/T ratios ....)
- What has been your experiences in applying these concepts to a total supply chain? Getting real "nitty/gritty", some Kodak issues are:
Our current guideline is to skip "theoretical" and equate entitlement to the "best-ever"/port to port time for the normal transportation mode.
That is, what if "entitlement" is beyond what a rational business case would support, considering design limits of high cost equipment (e.g. very small lot sizes may be "feasible" but would never be economic for some equipment; and there is not a business case considering inventory and customer benefit that could justify changing equipment)
The current response to that question is "find a way to make economical - that's why we hired a smart person like you." Surprisingly, there are some who do not find that a totally satisfying answer.
There are also issues around different batch sizes between functional operations, but I sense that everyone's eyes have long since glazed over so I'll stop with the first two issues.
The Optical Storage division of Kodak develops, manufactures (in low-volume),and sells high end optical storage devices. These included high speed CD writers and mass storage jukebox systems. There was a concern within the division that Kodak's procurement policies were not optimized for low-volume manufacturing. To test this hypothesis, a purchasing model was developed in an attempt to accurately assess cost differences between Kodak’s current procurement policy and an optimal buying policy outputted by the model. Although the model can be used for high volume buying decisions, it is especially designed for low-volume (under 1000 units annually) production environments.
The model calculates an optimal buying strategy based on a number of input parameters. These input parameters include supplier volume quotes (in step function form), the cost of capital, set-up and delivery costs, and ordering and inspection costs. By using graphical surface plots as output, the user is able to identify an optimal procurement policy for individual suppliers.
The first graph plots order quantity and the number of deliveries against unit cost. This provides an optimal delivery schedule and order size commitment (blanket PO) with a supplier, based on a fixed demand. The second graph plots order quantity and demand against unit cost while holding the number of deliveries fixed. This allows the user to determine the change in unit cost based on shifts in demand. Users of the purchasing model use both plots to help determine the lowest cost while minimizing risk.
Three scenarios were run to determine the expected cost reductions possible by using the purchasing model. A low risk, medium risk, and high risk strategy was employed with maximum supplier commitments of one, two, and four years respectively. Since low-volume production environments have less units to spread fixed costs over, it was important to consider commitments greater than one year. Current buying policies were entered into the purchasing model to determine actual costs and then compared with the optimal buying policies determined by the purchasing model for each scenario.
The low risk scenario resulted in an expected component cost reduction of 4.34% annually. The medium risk scenario’s calculated component cost reduction was 8.94% annually. The high risk scenario only achieved an 11.35% reduction in component costs annually. Clearly the risk of obsolescence, engineering changes, and demand variation increases with time, and each scenario.
By assembling a team of engineers, buyers, assembly operators, marketing personnel, and finance individuals, the model can be used to strategically buy most components, especially high cost components. The team can use its cross functional knowledge to determine which components are likely to change and when that change will occur. While four year commitments to suppliers seem risky, substantial cost reductions can be obtained around both low and medium risk scenarios.
Please contact me if anyone has any information or notices relevant to LFM masters students and the Scheduling and Logistics Group. Information from companies on possible internships, ideas from students on types of projects that would interest them etc.
This article describes research conducted by David Kletter as part of a recently completed thesis supervised by Professor Stephen Graves.
Motivated by applications at GM metal stamping plants, we study the planning and control of a single machine production/inventory system when the machine is unreliable. GM’s Lansing Auto division and GM’s R&D Center have supported this research and have been actively involved at each stage. It is our hope, however, that the research will have applicability outside of this particular application. There are two distinct parts to this research, which we describe separately.
The first part studies the problem of when (and how much) overtime to run on an unreliable machine. To motivate this problem, consider the following (intentionally oversimplified) example. Suppose we must deliver 500 units of our product to our customer tomorrow morning, but we only have 400 units in inventory, so we must manufacture 100 units. It is now 4:00 PM; there is one hour left in the work day; our machine is currently set up to produce this product; and the machine can produce 200 units per hour. Unfortunately the machine fails on average every 30 minutes and when it fails, requires 15 minutes on average to fix. If the machine did not fail, we could produce the 100 units in half an hour. However, due to machine failures, there is some probability that we will not be able to produce the 100 units by the deadline.
The production manager is now faced with several questions. What is the probability that we will be able to meet our demand? What is the expected shortfall? At 4:01 PM, the machine fails. Now what is the probability that we will be able to meet our demand? The production manager now considers using overtime. Suppose union rules dictate that plant management must decide by 4:30 PM if overtime will be run for one hour at a cost of $200. What should the decision be? Suppose instead that after running one hour of overtime, the production manager can stop overtime at any point. When should we stop? Suppose that we can delay shipping the product until noon if we pay $5 per unit extra for express freight shipping. How does this change the decision?
We develop mathematical models that assist a decision maker in answering questions such as the ones posed above. These models could be used as part of a manufacturing control system in a real manufacturing operation, and are capable of analyzing a multi-product scenario with multiple shipping deadlines and multiple overtime opportunities. One should envision these models embedded in a software tool that would receive data in real time from the shop floor and assist plant management in decision making.
The second part of our research compares and contrasts several different policies that could be used to control a single, unreliable machine producing multiple products to stock. Our goals are to obtain a better understanding of the strengths and weaknesses of different policies, and insight into how the policies that we consider compare against one another in different environments. We hope that our findings will assist decision makers in the selection of an operating policy that is best suited for a particular environment.
An operating policy specifies when the machine should produce (versus sitting idle), what product it should produce, and how long it should produce the current product. One such policy is the classic lot size / reorder point policy. This policy continuously monitors the inventory level until some "reorder point" is reached, authorizing production of that product. If at least one product is below its reorder point, the machine will not be idle. When a product is produced a fixed number of units are produced.
Another example of an operating policy is a cyclic sequence. In this policy, the parts are produced in a fixed sequence, such as 1-2-1-3. As part of the sequence, we might insert a fixed amount of idle time (e.g. after producing product 2, idle for an hour). When a product is about to be produced, the inventory level is observed and the production quantity is selected to bring the inventory position up to some predetermined level
In total, we compare seven such policies. We compare these policies in a manufacturing system with three primary sources of variability. The first is, as before, the unreliability of the machine, so that it is uncertain how long it will take to produce a given number of parts, or how many parts can be produced over a given time interval. The second source of variability is demand variability. The last is waiting for setup crews. In a metal stamping plant, a setup crew may be shared by several machines. As a result, a machine may or may not have to wait in order to be set up, and this will induce variability in the system.
We compare the policies using real data from two production lines at a General Motors metal stamping plant. For these lines, we find that a policies based on reorder point methods have relatively superior performance. A special variant that considers the availability of setup crews is also found to be effective.
In summary, we have developed models that focus on the planning and control of a single unreliable machine. We believe these models have both theoretical and practical value.
Jeannette Nymon, 412-337-4325/ -2737, nymon_jg@atc.alcoa.com
Joe Velez, 319-359-2163/ -2755, VELEZ01@ssw.alcoa.com
John Harrington, 508-880-8319, harrington@islnds.enet.dec.com
Stu Sharpe, 508-493-5059, sharpe@asabet.enet.dec.com
Bruce Arntzen, 508-493-5059
Bill Colwell, 313-322-0260/ -390-9232, usfmcmyh@ibmmail.com
Gene Coffman, 313-5692-2079/ -2381
Sita Bhaskaran, 810-986-2302/ -0574, sbhaskar@cmsa.gmr.com
Paul Williams, 415-857-3964/ -6278, williams@hpl.hp.com
Shailendra Jain, 415-857-3597/ -6278
Scott Elliot, 707-577-5545/ -2104, scott_elliot@hp-SantaClara-om2.0m.hp.cm
Steve Smith, 602-554-8066/ -6838, SSmith@FA.intel.COM (or stephen_p_smith@ccm.ch.intel.com)
Ron Caldwell, 716-726-2724/ -6945, Caldwell@Kodak.COM
George Daddis, 716-477-3241/-588-8274, KP26.567736@kodako.kodak.com
Marty Melone, 716-726-0352/ -1312, LOCKOVM1.247550@kodako.kodak.com
Chuck Petro, 716-726-3190/ -6945, LOCKOVM1.VAB811@kodako.kodak.com
Bryan Parks, 716-726-9382/ -????, LOCKOVM1.LL504729@kodako.kodak.com
Bryan Gilpin, 708-576-5470/ -538-7791, abg005@email.mot.com
Anantaram Balakrishnan, 617-253-0467, anant@mit.edu
Stan Gershwin, 617-253-2149, gershwin@mit.edu
Steve Graves, 617-253-6602, sgraves@mit.edu
Tom Magnanti, 617-253-6604, magnanti@mit.edu
Vien Nguyen, 617-253-0486, vien@mit.edu
Don Rosenfield, 617-253-1064, donrose@mit.edu
Larry Wein, 617-253-6697, lwein@mit.edu