July, 1996
Last March, when workers at two Dayton, Ohio, brake factories went on strike, the ensuing paralysis of General Motors plants across North America drove home a point that Stanley Gershwin has been trying to make for years: random disruptions, even minor ones, can propagate across a manufacturing system, causing massive headaches. The strike at the brake plants in protest against GM's practice of buying from nonunion sources caused the automaker's deliberately lean inventories of brakes to dry up quickly. More than 177,000 GM workers were laid off. Parts suppliers, idled by the shutdowns, laid off tens of thousands more. By the end of the 16-day strike, economists were putting the cost to U.S. productivity at $5-7 billion.
Gershwin, who is associate director of the MIT Laboratory for Manufacturing and Productivity, doesn't fault GM for keeping inventories low. Strikes, he says, are rare events that manufacturers can't necessarily plan for. But the nasty business in Autoland illustrates the value of adequate buffers. "Buffers are like shock absorbers," says Gershwin. "When there's a disruption in producing one part, they prevent the damage from spreading to other areas of a production process."
Exactly what level of inventory is sensible for a given manufacturing operation is a complex issue. In fact, Gershwin says, the whole question of how to compensate for disruptions is one that baffles many manufacturers. But just as there are ways to minimize disruptions (by boosting routine maintenance, for example), companies can take steps to keep a factory humming when random events do happen when machines break down, workers get sick, or parts fail to arrive.
Gershwin and his colleague Mitchell Burman, a recently graduated PhD student in Gershwin's research group within MIT's Leaders for Manufacturing Program, believe that difficulty in coping with routine disruptions arises because factory design tends to be less rigorous than other branches of engineering. "Certainly factories are complicated," says Gershwin, "but no more so than other complicated things like computers or automobiles or airplanes. Somehow a culture has developed that views factories as different, as much more intractable kinds of systems." As a result, he says, factory designers rely heavily on intuition. "But that intuition is not guided by a solid study of science the way an engineer's intuition about the behavior of fluids is based on 200 years of fluid mechanics."
Fads and Foibles
Without a firm foundation of understanding, manufacturers often misapply various fads or theories that are meant to boost efficiency. Consider the wildly popular "just in time" concept. The theory's emphasis on cutting inventory does save space and allow manufacturers to respond faster to market demands, according to the MIT researchers. The danger lies in taking the concept too literally. "Reducing inventory to nothing can cause more damage than having too much inventory," says Burman. Decisions about how much inventory each machine on an assembly line needs are complicated by tradeoffs between the cost of inventory and the cost of delays. Burman has developed a computer algorithm to weigh such tradeoffs, and now heads a consulting firm, Analytics, Inc., in Chestnut Hill, Mass., that helps companies sort through these and other production issues. As a crude rule of thumb, however, Gershwin suggests that the ideal amount of inventory is proportional to the length of the most common disruptions: if the breakdown of a certain trouble-prone machine usually takes three hours to fix, an assembly line should have three hours' worth of inventory to avoid grinding to a halt.
Disruptions are felt most keenly when companies succumb to another misconception: higher utilization is always better. "People think that once you've invested all this money in capital equipment, the factory should be run at 100 percent of capacity," says Burman. But suppose a company's promises to customers assume that products will be turned out at 100 percent of capacity, and a machine outage causes production to fall to 90 percent during a certain week. Not only will 10 percent of the customers be unhappy that week, but the plant will be 10 percent behind indefinitely. "Because you can't exceed capacity, you can never catch up," says Burman. Gershwin concludes: "You need a strategic amount of extra capacity to serve as a buffer, just as you need a strategic amount of inventory."
Another risky proposition, the researchers say, is the idea that each step in an assembly process should take the same amount of time. Manufacturers often try to "balance" their production lines, adjusting the speed of each operation in an effort to keep parts moving like clockwork from one machine to the next. A good idea in principle, says Burman, "but it's based on the assumption that nothing ever fails and everything moves perfectly." Each time a machine unaccountably speeds up or slows down, a bottleneck occurs. In a process with a large number of steps, "you can have bottlenecks bouncing all over the place." A factory would need extremely sophisticated logic to track, and adequately compensate for, all these bottlenecks.
A simpler solution, according to Burman, is to identify the slowest machine, the one most likely to create a bottleneck, and let it serve as the "gate" that controls the flow of the assembly line. "This way," he says, "you have one traffic light on your street instead of fifty, and you can more reliably tell customers when you expect to get products out."
Ironically, one trend that can worsen disruptions is the growth of automation. "You can't simply buy more computers and more robots and expect better performance," says Gershwin. A company will often replace an old, slow machine with its gleaming high-tech counterpart, yet fail to take into account the consequences of a breakdown. While a relatively simple machine can usually be fixed quickly and locally, Gershwin says, repairing a sophisticated computer-controlled machine may cost a day or more in downtime. "By investing in this high-tech machine, you've forced yourself either to put larger buffers around or to accept a lower production rate."
Automation can also go awry when robots are assigned tasks that are more accurately performed by humans. Building a robot that can adapt to irregularities is both difficult and expensive, says Gershwin. Toyota, for example, recently gave up on the idea of using robotic systems for the tricky task of installing engine blocks. A simple power-assist device controlled by a worker turned out to be a more reliable and flexible way of positioning engines in car bodies. Despite such pitfalls, the researchers are quick to affirm that they are not against high-tech manufacturing; they just want factories to be more aware of the need to take into account the likelihood of disruptions. "We're not Luddites," says Gershwin. "What we're saying is that you have to take a sophisticated and careful view of these things, and think out the consequences."
David Brittan
Stanley Gershwin (left) and Mitchell Burman are convinced that
factories would be more efficient if they were engineered as carefully
as other complex systems such as computers.
Reprinted with permission of Technology Review.