The players are based on work stations, or nodes. They do not move; they have agents to move the disks. Disks arrive to an input buffer, the players inspect the disks and the special symbols attached to them. They select disks for processing: paint them with different colors, carve and polish them and put them into an output buffer. Each time a player completes some work, or an agent moves a disk, the process is transformed. It moves from one state to another. A series of snapshots can show how the process evolves. It is a fact of life that most of the time the disks are waiting to be processed, and there is a natural tendency for bottlenecks to develop.

Decisions are decentralized and are made at least on two levels. Global decisions (plans), made by a Master of Ceremonies, MC, who represents a group of decision makers, specifies entering the disks on the far left, and the schedules. Local decisions are made by players, who dispatch and sequence the disks, and select which disk to work on.

"Scheduling" or Controlling Stochastic Processes

In the theory of stochastic processes it is customary to speak of control by feedback signals. In this parlance, the MC develops schedules and transmits them to the players as signals. The players use these signals to facilitate and guide their decision making.

Consider, for example, MRP, whatever version. Each disk has a schedule date written on it. The player considers this data item as an input to select a disk to work on. The player has a myopic view of the work station, and some of the neighboring stations. The player cannot see far up- or downstream and cannot tell how the decision will influence the global objectives of the game. The input signal should help. Bob Millard aptly describes the traditional interpretation of the date:

...MRP might tell you that you need to start some operation prior to today's date...which is very difficult to do.... [T]his is made possible only through the use of time machines, which so far exist only in the movies. (Millard 1996)

The theory of stochastic processes interprets the signals in a different way. Suppose the player needs to decide between two alternatives: (1) the scheduling signal is a week prior to today's date; (2) the signal is a week ahead. The player does not laugh, does not throw hands up; she takes the signals calmly and works on the first alternatives. In this framework scheduling dates have a different meaning.

Assume now, that the MC knows nothing about the control of stochastic processes, and wants to "schedule" production. The MC knows from past experience that when an order is released, it takes weeks to produce the parts, even if actual production takes only a few hours. Most of the time is taken up by doing nothing waiting. The MC develops system RPMXYZI, which puts together a structure of lead-times and releases the "schedule" to the shop. The MC lives in this artificial world, flies blind and tries to control a process without observing it. RPMXYZI is an open loop system. After a period of time the artificial world of RPMXYZI is shattered. The MC observes that the floor pays little attention to the system: items are made late and so are end items. She observes that there is not enough capacity. So the MC analyzes shop capacities, includes them in RPMXYZII and uses longer lead-times. Inventories go up, but the important thing is that schedules are better met.

But things are still not good enough; goals are not met. The MC now realizes that it would help to know what is taking place on the shop floor; it would be better to have feedback, a closed loop system. So she introduces a listening device that gives her some information despite the deafening bedlam. She develops RPMXYZIII, controlling a stochastic process, albeit only partially observed. Her artificial world is made more realistic by including Bayesian rules of decision making.

RPMXYZIV is a system that realizes that keeping schedules is only a part of the game. Many other metrics like profit, sales, cash flow, present values, inventory levels, work-force fluctuations, must all be considered. Thus lead-times are adjusted.

RPMXYZV is a system that is interpreted as a set of control signals to the players. Lead-times and the logic of RPMXYZV are decision parameters to be used in the calculations. Partial knowledge of the state of the system is used as feedback when calculating the signals. Decision-making is still decentralized, players still make local decisions, and the MC does not get involved in them. Players use heuristics to make local decisions. Hopefully, global goals of the firm will be better met. But the MC cannot hope that things will move on "schedule." The performance of stochastic processes is measured by probability distributions of the various metrics.

RPMXYZC, the ultimate planning, controlling and scheduling system, will be on the Intranet and the MC will manage the databases, do the calculations and provide expert systems to facilitate decision making on the local level. RPMXYZC is a dynamic, sequential, cybernetic system. The parameters can be changed, WHAT-IF questions can be answered by simulation. The system is frequently updated, improved, and adjusted to the rapidly varying environment.

Conclusions

When such a framework is established, it becomes possible to present and critically review the many, currently advocated Alphabet systems, and fit them into a general approach to production and operations management.

Reference