R. Morley Inc

Articles:


Click On The Black and White Barn Picture To Get Large Color Image.


Chaos and the IS executive
Christopher Meyer
05/20/96

While complexity theory is far from a mainstream management tool, some brave souls are evaluating how the new science can be applied to their businesses. Companies such as General Motors Corp., Citicorp, Swiss Bank Corp. and Deere & Co. have even built applications using complexity theory that help to cut costs and create value. But for the science to take hold in corporate America, CIOs and their IS departments must become more involved.

Complexity theory starts with a core idea that simple objects, or agents, can interact to create elaborate and unexpected behavior. Take the early model called "boids," developed by computer animator Craig Reynolds in 1987, which simulates the flocking behavior of birds.

In Reynolds' model, each boid is represented by a wing-flapping symbol that is programmed to obey three simple rules: 1) fly in the direction of the other boids, 2) try to match velocity with neighboring boids and 3) avoid bumping into things. When the simulation begins, the boids are positioned randomly in space, yet they soon form a flock that wheels and turns and reacts realistically to obstacles placed in its path, much like a flock of birds.

The tendency of the boids to flock is called emergent behavior. The turning of the flock is not designed; it emerges when the boids, or agents, interact with each other and their environment. This behavior, not of each bird but of the flock, was accurately modeled through simulation.

Some corporations are using agents, emergent behavior and other concepts of complexity theory to improve inventory control, logistics management and production scheduling. The resulting productivity gains and cost savings are often dramatic.

Consider GM's Fort Wayne, Ind., truck manufacturing facility. In the early '90s, Dick Morley, inventor of the programmable logic controller in the late 1960s and now an advocate of complexity-based computing solutions, dismantled the factory's command-and-control system for coordinating painting booths. He then designated each booth as an independent agent capable of "bidding" on new paint jobs, depending on the booth's ability to perform the work quickly and at low cost.

This new, self-organizing paint system reduced color changeovers at paint booths by 50% and saved GM $1 million in paint costs alone.

Although Morley's system is elegant and effective, IS personnel at GM had limited involvement in the project. Morley's project team consulted with a core group of GM engineers and software specialists - many of whom were very skeptical of the paint booth work - but it was Morley's firm, Flavors Technology, that orchestrated the implementation.

If GM's small example of complexity theory in action, and the IS overhaul that goes along with it, can yield such positive results, imagine the impact CIOs could have in championing such practices in their own companies.

Deere's natural selection

There are any number of variations on the complexity theme. Bill Fulkerson, a staff analyst at Deere's Moline, Ill., engineering labs, solved the farm equipment maker's scheduling difficulties using a survival-of-the-fittest complexity-based approach.

At the Moline factory, the variety of customized planting equipment being assembled - 1.6 million possible unique models in all - makes generating an optimal production plan beyond the capacity of any traditional expert system, let alone any human. In the past, untold hours were eaten up in shuffling data to hammer out weekly manufacturing schedules that rarely came close to production targets.

Fulkerson, with the aid of Cambridge, Mass., software developer Optimax Systems Corp., resorted to using a "genetic algorithm" to produce the schedule. Genetic algorithm software, originally developed by professor John Holland at the University of Michigan, formalizes the process of natural selection to allow solutions to emerge, rather than be calculated.

Each night at Deere, a single PC on the factory floor downloads data from the plant's database and generates an initial set of trial schedules. These schedules are then permitted to "breed," or combine, to create new and improved schedules. Each successive schedule is tested for fitness - whether it produces a more efficient throughput than the others generated during that iteration - and the fittest schedules are selected to breed with each other and produce new generations. Evolving solutions, rather than engineering them, is the essence of a complex adaptive system (CAS).

In Deere's case, roughly 600,000 offspring schedules are generated and tested overnight; the next morning's result is a schedule that's not perfectly optimal but comes very close. No worker is involved in producing the resulting work plan, other than turning on the computer and picking up the output. The solution is a product of evolution, not engineering.

Deere's results have been significant. Overtime at the Moline plant has been drastically reduced, while monthly production figures have increased.

As in GM's case, IS people were only peripherally involved with Deere's trial implementation of the scheduling system; Fulkerson's team and Optimax Systems pursued it as a side project that didn't attract much corporate interest until it started bearing fruit.

The company now plans to roll out similar implementations at other facilities, and IS professionals will have to make sure that factory systems' infrastructures can support the automatic, real-time data gathering required by the genetic algorithm approach.

Chaos at Cemex

While Deere and GM have leveraged CAS at the factory level, Cemex SA, the $3 billion Mexican cement and concrete company, is reworking its entire delivery system around complexity theory.

The problems Cemex faces in delivering ready-mix concrete to its customers involve so many constantly changing variables that scheduling defies linear, command-and-control optimization. Weather and capricious government site inspections can destroy a construction site's work plan in minutes. Work stoppages, due to lack of funds and materials, are frequent. Congested traffic in many Mexican cities makes matters worse, as ready-mix concrete is specially formulated for each job and has a short window in which it can be poured to achieve maximum effectiveness. On average, 50% of Cemex's orders are canceled on the day of delivery, and many others are rescheduled.

Imagine keeping track of 20 to 300 trucks within a service area such as Guadalajara or Mexico City and you have a sense of the mayhem Cemex deals with every day. "It's like being an oil tanker in the middle of a hurricane instead of a surfer," jokes Ken Massey, director of Cemex's Center for Business Processes.

With the help of a CAS approach, Massey's team took less than a year to transform Cemex's ready-mix concrete delivery system in Guadalajara from a tanker into a surfer. "The surfer has a great time riding the waves, while the tanker fights for survival," Massey says.

By combining revamped expert systems with global positioning satellite equipment, the new Cemex system can pinpoint the location of every concrete truck in close to real time and at very low cost. Manipulation of this data and other information about truck cargo, plant conditions and construction sites allows trucks to become independent agents within a network of plants, rather than appendages of a particular plant and held to brittle, centrally controlled schedules.

The delivery system can now react much more quickly to rerouting problems, thus freeing staff to focus on problems requiring human intervention, such as negotiating essential shipments of concrete ingredients.

For example, when the Cemex system receives changes in construction plans from a customer, the system can instantly identify orders for similar concrete formulas and reroute the appropriate trucks to plug emerging gaps in order fulfillment. Also, the system can divert trucks to nearby plants and have their formulas modified, thus fulfilling orders with trucks originally intended for other destinations.

These delivery system capabilities have allowed productivity of the mobile equipment to increase by 30%, or twice what was originally expected. Equipment maintenance costs per ton of concrete have dropped dramatically, as has fuel consumption. New orders are no longer lost because customers can get through to a phone system that had been previously tied up with truck-routing calls. Business is more evenly distributed across plants, and loading times have decreased accordingly.

The bottom line to the customer is that Cemex's delivery window has dropped from the industry average of a couple of hours to within 30 minutes, and the company is moving toward a 10-minute window.

Although Massey's team spearheaded the complexity implementation, IS professionals were involved throughout the process. Cemex's computer systems need a steady flow of real-time data to make the new delivery scheme work - a requirement that can't be met without a responsive information technology infrastructure. Cemex is looking to spread the use of this and other complexity applications into other parts of the organization, and the IS department will be needed at every step.

GE's solution

Another challenging optimization problem involves product design. General Electric Co. used genetic algorithms as part of a hybrid artificial intelligence solution to improve bottom-line performance in the design of the Boeing 777's jet engines.

While addressing stringent design specifications for the Boeing aircraft, GE engineers determined that simplifying the engine's compressor would go a long way toward meeting design goals. However, the compressor already had been refined to the point that human designers were struggling to find any incremental improvements.

With deadlines fast approaching, project managers opted to use an engineering software package developed by GE's research and development lab. The package, called Engeneous, allows a design to evolve to improve its match with a large number of constraints.

After a number of critical design factors were input, one simple but profound rule was introduced: Design a compressor with only six stages, instead of the previous seven. "At this point, we were just looking for new ideas and were curious to see what Engeneous would produce," says Peter Finnigan, a research manager on the project. As it turns out, the software produced very powerful results.

Engeneous coded each design factor as a "digital chromosome," then mixed these chromosomes together to form trial designs. After a breeding and fitness testing process similar to the Deere scheduling program, GE had a viable six-stage compressor design in less than a week.

Systems like Engeneous often help refine designs that have been reworked heavily by humans, Finnigan notes, but the compressor solution yielded additional efficiencies on top of the design criteria. For instance, the six-stage design required less metal, which meant less weight and, thus, a greater-than-anticipated decrease in fuel consumption.

Of biodiversity and whirlpools

Complexity theory's usefulness is not limited to the world of manufacturing; it also has been embraced by the financial services industry. For example, Citicorp's new complexity-based approach to handling portfolios is an application of biodiversity to portfolio management.

Through years of experience, Citicorp has developed risk-management algorithms for swapping currencies and around-the-clock trading. These single-best solutions could be applied across all of Citicorp's investment portfolios, but the firm won't allow an across-the-board application of one procedure. That would create a "monoculture," a basket of portfolios susceptible to the same external shocks. Instead, Citicorp tries to maintain diverse near-best approaches to risk management, thereby protecting itself from the exposure inherent in a monoculture.

With the aid of complexity theory, the firm has developed multiple risk-management algorithms that constitute a healthy ecosystem for its portfolios, safe against any one effective attacker.

Another application of complexity theory in financial services is to predict the unpredictable. For example, although CAS cannot predict the appearance of a whirlpool in a stream, it can help to identify the conditions that create the whirlpool and the likelihood, based on those conditions, of its persistence for a given period.

Finding the conditions that create financial whirlpools is allowing Swiss Bank to exploit periods of turbulence in financial markets by recognizing patterns of persistent factors. Building on basic work done by The Prediction Co., based in Santa Fe, N.M., and founded by chaos researchers Doyne Farmer and Norm Packard, Swiss Bank has created a system that allows it to rapidly identify the development of a turbulent situation in the market, make the right moves, then stop making those moves before they start to produce losses.

Other investment firms, too, are experimenting with the use of complexity-based systems to gain faster and more in-depth comprehension of seemingly chaotic market forces.

What's next

Successful business applications make it clear that chaos, or complexity theory, is not simply the stuff of ivory-tower musings. The new sciences, and the new mind-sets they entail, are helping hard-nosed managers to derive real, measurable business-performance improvements.

But it's still the early days. Inroads in highly analytical, computational fields such as engineering and financial modeling undoubtedly will continue, but other applications such as digital organisms - recently developed by Excalibur Technologies Corp. to recognize fingerprints, faces and Japanese kanji characters - could someday be employed in resume scanning by human resource departments, allowing companies to more efficiently and accurately match job openings with prospective candidates. Also, pattern recognition technology like that used by Swiss Bank may direct more precisely targeted marketing campaigns.

Indeed, someday, we may be able to model high-performing organizations in terms of the self-organizing behavior of individual employee "agents" and the environmental factors to which they adapt.

Further in the future, we may come to see CAS not as a tool kit for operations management, but as an element of a new paradigm, with evolution succeeding design as the model for managing complex systems. The increasing connectedness of the "econosphere" makes economic agents, such as the trucks at Cemex, elements of an evolving organism, rather than cogs in a static machine. Perhaps the machine metaphor for business has reached the limit of its usefulness, as businesses, by becoming more connected, become more complex.

The CIO's role in complexity theory

At some corporations that have applied complexity theory to real-world business problems, the IS department is involved only peripherally. Here are a few tips for IS leaders who want to raise their awareness of - and their profiles in - the application of complexity theory.

1. Get comfortable with the mind-set.

The first hurdle to tapping the energy of self-organizing, adaptive systems is to acknowledge that top-down, dictatorial control is not only inadvisable, it's also impossible. Instead, the role of the executive is to enable the organization to proceed by establishing the basic framework of rules and goals and then to recognize and support good solutions as they emerge.

"The logic behind chaos theory flies in the face of the education and training [CIOs have] received," remarks Ernie Vahala, retired group director of manufacturing engineering for General Motors Corp.'s truck division, who worked with Dick Morley on GM's paint booth project. "However, if they participate in the design of a complexity-based system and see how well it works, they tend to become converts," he says.

2. Learn about the theory and tools.

In order to get people to buy into complexity, IS leaders have to know enough about it to present its concepts to management in a nonthreatening manner. As with all management problems, the devil ultimately is in the details, and any work done now to understand the mechanics of tools and techniques such as genetic algorithms and simulation modeling will be well invested.

3. Keep your eyes open for that killer app.

The real payoff for an IS leader - and, therefore, the real vehicle for popularizing complexity within your organization - will come when you are the one to step forward with a complexity-based approach that promises and delivers better performance.

Scouting for those killer applications is not such a hunt in the dark when the search starts with a basic question: What problems in this business are crying out for optimal solutions but have too many conflicting constraints to be optimized?

 

Additional reading

"Chaos Inc." by Simon Caulkin, in Across the Board, pages 32-36, July 1995.
At Home in the Universe: The Search for the Laws of Self-Organization and Complexity, by Stuart Kauffman, Oxford University Press, New York, 1995.
Out of Control: The Rise of Neo-Biological Civilization, by Kevin Kelly, Addison-Wesley Publishing Co., Reading, Mass., 1994.
Hidden Order: How Adaptation Builds Complexity, by John Holland, Addison-Wesley Publishing Co., Reading, Mass., 1995.

Symposium: Dick Morley's Chaos in Manufacturing Conference. Talk to people who have used complexity theory successfully in the business world. Contact R. Morley, Inc.


For e-mail, please send to Morley@barn.org

Homepage
Staff - Table of Contents - Guest Book
FAQ - Links - Links (Text Only)
 

Send e-mail to rmi.info@barn.org for more information.
Please send e-mail to webmaster@barn.org regarding web site structure.
Copyright © 1996 - 2002, R.Morley Inc. All Rights Reserved

R. Morley Incorporated
614 Nashua Street, Suite 56
Milford, NH 03055-4992 USA
Tel: 603-878-4365 FAX: 603-878-4385