Using Genetic Algorithms to Conquer Tough Process Problems
Don C. Johnston, Vice President, Roxtar Consulting, Inc., Titusville, FL, USA
Keywords: Optimization, Flow, Genetic Algorithms
Traditional Lean Six Sigma (LSS) tools help us improve processes dramatically; but for highly complex problems, good improvements may fall far short of what’s truly possible, “leaving money on the table.”
Tools like DOE and response surface methods can be impractical when the problem at hand reaches a certain level of complexity. Scientists and engineers are increasingly using Genetic Algorithms (GAs) for such intractable problems where traditional methods struggle.
Mimicking the basic mechanisms of animal genetics (selection, gene crossover, and mutation), GAs are helping professionals solve tough challenges in a myriad of fields. We in LSS should place GAs in our toolbox as well. Because of their potential utility, this session will explore how GAs can be applied to difficult problems that have exponentially large numbers of possible solutions.
Facility layout is an example of such a large search space. Consider placing a piece of equipment on the factory floor: How close to the shipping door or the main aisle should the machine be? What about proximity to power? How should the machine be oriented? How close should it be to the previous workstation in the process? Now, consider how many workstations are in the process, each with the above variables to resolve. And perhaps there are many processes being performed in the facility simultaneously, sharing equipment; the number of possible configurations explodes. How can we find the best design?
While Lean techniques give us good guidelines for such layouts, we may be missing out on huge savings because guidelines and best-practices have limitations, such as their scope. The same situation is true for many process problems, not just facility layout. Consider: What is the optimal way to schedule work in an organization? How do I arrange arrival and departure of trucks to receiving docks? What’s the optimal way to store items in a warehouse? How should I hire and schedule people to maximize output? All these situations have potentially innumerable configurations and solutions. Choosing the right solution among so many can mean the difference between profit and loss.
While we may employ traditional analytical solutions to these problem domains, such approaches may require oversimplification of the problem, leaving better solutions undiscovered. Genetic Algorithms, in contrast, allow us to search through mammoth solution spaces to find results that outperform traditional analysis.
Imitating human genetics and heredity, GAs can produce optimal or nearly optimal solutions in such huge solution spaces. GAs are relatively easy to program; many existing programming libraries can be used to quickly model various problems. GAs have been used successfully in many fields of engineering and science to produce optimal designs, proving their utility.
This talk will explain the basics of the genetic algorithm and how it can be used in LSS. The speaker will provide examples of GA use in solving process improvement problems. Attendees will walk away with a better understanding of how this technology can be applied to their own process problems.