Stanford University Home

Stanford News Archive

Stanford Report, August 9, 2000

The end of the line for the assembly line?: 7/00


You've just stopped for gas before heading out on the highway for your commute. One of your tires has a slow leak and is treacherously close to going flat, but you don't know it.

You shove your credit card into the slot and wait for the display to tell you to start pumping. Instead, it tells you that you'd better put some air in your tire, pronto -- also, that an engine bolt is a little loose (nothing serious) and, by the way, did you know it's almost time for your 30,000 mile checkup? Oh, one more thing: There's a Federal Express packet waiting for you. Where will you be this afternoon around 2 p.m., the gas pump wishes to know.

Ford factory, first moving assembly line, 1913, Highland Avenue, Detroit, MI
(Courtesy of the Frances Loeb Library, Graduate School of Design, Harvard University)

Believe it or not, the technology that will make such exchanges possible is already being put in place as you read this. A world in which machines routinely talk to one another is closer -- perhaps much closer -- than we may think, several experts agreed at a recent workshop held at on campus.

Organized by research assistant Miguel Pinilla of the Department of Mechanical Engineering, the workshop -- imaginatively titled "The Factory Central Nervous System" -- brought together participants with an interest in the burgeoning role that information systems are playing in optimizing design and production in modern factory environments. The workshop was sponsored by the Alliance for Innovative Manufacturing at Stanford (AIMS), a campus-based joint venture of the Graduate School of Business, the School of Engineering and several corporate partners whose mission is to encourage advances in manufacturing and to disseminate these breakthroughs throughout industry and academia.

Thanks to advances in data telecommunications technology and to the ubiquity of tiny sensors and memory chips (what Pinilla and others in the field call "embedded intelligence") that can be implanted in electrical and mechanical contraptions of all varieties, interactive appliances may become as taken for granted tomorrow as our ability to pay for our gas without talking to another human is today. This revolution in the way inanimate objects interact also may spell the death of traditional factory assembly lines and the birth of a radical new approach to manufacturing.

Sun Microsystems, of Palo Alto, has produced the communications-oriented programming language known as Java and its close cousin, Jini, which allows appliances with embedded intelligence to exchange data with one another. According to workshop participant Robert Atherton, Sun's worldwide manager for process control, Sun now is working with the Big Three automakers, three international oil giants and a few other companies to expand Java-enabled communications capabilities to gas pumps and cars.

The automotive and gasoline-refining industries, combined, comprise a market with annual revenues on the order of $5 trillion. "There are about 250,000 gas stations worldwide," Atherton noted, "and you go to one of them maybe once every eight days, much more often than you see a mechanic." Eventually, he added, cars will be self-driving, so we may be spending more time in our cars than we do already. But well before that time, he said, "your car will be able to talk to diagnostic equipment to say, 'Here's what I'm made of, and here's my history.'"

Benefits will flow in both directions. In exchange for the obvious advantage to drivers of such things as automatic maintenance monitoring, discounts for frequent patronage (or coupons for goods from co-sponsoring retailers) and notification of impending events, the companies sponsoring the communications will be able to capture consumer information.

Embedding self-diagnostic chips in any significant proportion of a car's components -- some 15,000 to 20,000 of them at last count -- will introduce yet another layer of complexity into a manufacturing process that is already mind-bogglingly intricate. Computers, airplanes and other high-tech products are every bit as complex. It currently takes about 600 engineers to design a computer chip, said participant Raju Mattikali, a Boeing engineer; the next generation will require more like 1,500. Yet, Mattikali said, a chip is by and large a stack of two-dimensional arrays and is typically tossed out when it stops working, whereas the network of tubing that connects components on an airplane is fully three-dimensional ("like a mess of spaghetti") and must accommodate moving parts and remain reachable for repair.

Yet with all this complexity, product cycles -- the windows of time before which goods become obsolete and unsalable -- continue to shrink, while customers demand ever-faster delivery on the orders they place, participants agreed.

This places a premium on getting a product assembled and out of the factory door without delay. And it is here, said Pinilla, that the concept of machines talking to one another (as per the smart car, etc.) yields a radical revision in manufacturing practices: the demise of the assembly line.

Much delay in manufacturing results from queuing, or the lining up of partially assembled products in front of the machine that will perform the next operation, Pinilla explained. And this in turn is the result of what we call "the assembly line": a rigid scheduling of a part's trajectory from one machine to the next.

"Suppose two separate customers walk into a Starbucks," said Pinilla. "The second customer waits while the first customer pays at the register; meanwhile, the coffee machine attendant is idle until the cashier shouts out the order. Then, after paying, Customer Number Two waits again at the coffee machine because Customer Number One's espresso is still a work-in-progress. But suppose instead that one customer were to go first to the register, the other directly to the coffee machine. The first tells the cashier what he or she wants, and is charged for it; the second places an order directly with the coffee machine attendant. Then they trade places. You reduce the waiting time, using no additional resources."

Pinilla's research has shown that by moving from the single-pathway assembly-line concept to a choice of possible machine-to-machine trajectories, a manufacturer incurs fewer bottlenecks, increases utilization rates of machines and speeds production.

But how does a part identify itself as ready to be worked on? How does the appropriate machine "know" how to signal its availability? How does the part "know" where to go when it gets that signal? The answer, Pinilla said, is to introduce the interactivity that lets inanimate objects communicate into the womb (the factory), allowing "smart parts" to communicate with smart machines.

"By doing only that [without adding any new capacity], you can reduce the ratio of time a part waits in line to the time it's actually being worked on by up to 80 percent," because if a machine is available the part does not have to wait -- it goes to the first available machine. "I don't expect you will get 80 percent in the real world, but you can expect big improvements. How much improvement depends on how much disorder you can tolerate."

One virtue of the old-fashioned, choiceless assembly line is precisely that: There are no decisions to make. In contrast, the jettisoning of such a centrally planned, fixed, single-assembly path and the adoption of a multiplicity of pathways to be determined by real-time factory-floor conditions lead to competition among machines for parts, and vice versa. To resolve these conflicts, said Pinilla, imagine "market-based control" protocols: "What if we set the parts loose in the factory, and let them find their optimal path? The jobs-in-progress bid for machines' services with price: 'I can pay more because the company that ordered me will pay extra for early delivery, and I need a hole punched in me and some polishing around my edges. Right now!' Machines bid for jobs with cost: 'I am a milling machine. I can do milling. I cost so much an hour and I can work this fast.' If two machines become available at the same time, the one with the lowest cost wins."

Thus, the staid assembly line becomes a continuing multitiered auction, as machines offer their services to parts awaiting assembly even as those very parts bid against one another to be at the front of the line. In an age of outsourcing, the concept can be extended beyond the factory's brick walls, Pinilla said. "Get your vendors to program their machines to talk market protocol, plug into your network and bid for jobs, and let your suppliers become part of your factory."

The workshop participants predicted many nonfactory applications of appliances universally talking to each other in addition to garrulous gas pumps and complaining cars. On the plus side, if you set your alarm for 10 minutes earlier, your coffeemaker will automatically start up 10 minutes earlier. But then, there's also the potential for a rolling international toaster strike right around breakfast time, depending on your time zone. SR