The Payoff of Manufacturing Process Simulation

The first time you hear about simulations they sound like magic from a fairy tale: a technology that shows you alternative futures and allows you to pick the one you like best before that future arrives.

It is hard to imagine anything more useful, especially in business. What will happen to your maintenance budget if you run the production line at 110 percent for 24 hours, or 115 percent for 12 hours? If machine X breaks, is it better to shift to Y or Z or some combination of the two? Suppose you need to recall a product: How will your supply chain work in reverse? Want to know whether a phase of planned construction will block access to your store? Just set up the simulation, dial ahead to the dates of concern, and see for yourself.

Simulations are not impossible, but they are enormously expensive. Invented more than a hundred years ago, the technology has been confined to the military for most of its history, because only militaries could pay the bill. Before computers, the total was calculated by the cost of carving huge gaps out of the schedules of dozens of high-ranking officers so that they could sit for days around a table, pushing counters over a map.

Computers were able to lower the barriers of entry a little but not as much as you might think. Simulations calculate how a set of interacting objects (consumers, trucks, factories) changes each other’s attributes or states (location, price, production rate) during a given period. Each output feeds back through the underlying equations to arrive at another set of changes, over and over, for as far ahead as you like. Typically each successive update cycle or “tick” takes exponentially more resources. If you have 10 interacting objects in your simulation, each able to occupy 10 states or values, generating the first update cycle will require 100 calculations; the second, 10,000; the third, a million and so on. If each tick represents a 24-hour day, all the computers in the world would not suffice to represent the state of affairs in your simulation after a week?and 10 objects with 10 states each is a very small world. That huge volume of choices can be whittled down by clever programming, but that doesn’t come cheap either.

As intractable as that picture might seem, during the 1980s the gradual accumulation of experience and the accelerating empowerment of Moore’s Law forced the door to the technology open, bit by bit. By the early ’90s, the technology had spread enough for us to run a story on it (“Some Simulating Experiences,” Nov. 1, 1993). We were especially interested in a recent innovation that both illustrated and facilitated the spread of the technology: the use of animation as an interface. Animation lets users view output directly, as icons visibly moving in an environment, instead of inferring it from a table of numbers. Animation moved business simulations out of the analyst priesthood and into a tool that could be used routinely.

However, they were still not used very often. The huge resource demands, most of which lived in the programming end, meant that companies reserved the technology for high-end problems such as strategic planning. Applications that could pay the bills and had turnarounds measured in days were acceptable. The biggest payoff?operations?was still off-limits. If simulations could handle day-to-day matters, users could add arbitrary amounts of intelligence to every step in the production process. Questions that managers just guess at now could be answered systematically. But for a system to be feasible, answers needed to come back in minutes?not hours, let alone days.

During the past few years, a raft of new tools has brought this future closer. One example is the development of industry-specific simulation templates that allow nonspecialists to build simulations customized to their own company in a few hours, practically on a point-and-click basis. According to David Kelton, professor of management science at Penn State University, the templates have made it possible to move simulation technology from large manufacturing companies to service industries: for instance, using a call center template to build a simulation that in turn can determine staffing levels for a particular company.

These improvements have begun to breach the cost barriers that kept simulations out of operations. Rich Ryan, president of Rockwell Software, an automation software vendor based in Milwaukee, says that in his experience operational simulations are used most often as scheduling and setup tools. For instance, carpet production involves a long list of dye applications that can interact in complicated ways. Powerful simulations have proved useful for working through the scheduling complexities involved. Ryan also suspects that the growth of outsourcing is spreading the use of simulations?both directly, by making it easier to support the talent needed for a specialized simulation shop, and indirectly, by giving managers more options to juggle.

Simulations have become one of those technologies that tick along in the background, incrementally improving year by year but never quite getting nominated as “the next big thing.” However, they are potentially significant on several levels. Managers can optimize?in a systematic way?production issues that hitherto were dealt with by intuition. They can explore otherwise forbiddingly complicated environments, such as exotic supply chain configurations. And these decisions can be justified in the most convincing possible way: with little pictures.

In the mid to long term, such simulations might even permit the automation of middle management. Up until now there has not been anything like business artificial intelligence?machines that can look at a situation and make the kind of inferences an experienced human manager would about what needs to be done. With simulations, this deficiency doesn’t matter?the computers can just run the software and see directly which choices will work and which will not.

That scenario may not be imminent, but there is no obvious roof to the degree of improvement for simulations. As long as Moore’s Law rolls on, simulations will get more powerful and capable, reaching “down” into deeper levels of detail and “up” into more management issues and larger regions of the supply chain. Visibly or not, they will make the productive cycle steadily more efficient every few years for generations to come.