Self Organization in Design

Self Organization in Design

Author: Bart R. Meijer

Abstract: Principles of self organization are discussed as a frame of reference and a source of ideas for new design processes that can deal with more complexity in less time. It is demonstrated that set-based concurrent engineering makes effective use of these principles. Taking this idea one step further, an evolutionary organization for design processes is proposed.

Most academic institutions still teach structured design to their students Not because it is the best method, guaranteed to lead to good designs, but merely because it addresses all the relevant areas of design processes in a comprehensible way to students unaware of their own early design experiences.

Axiomatic design is not fundamentally different from structured design. The design phases are roughly identical. Axiomatic design is characterized by the design matrices, which represent an efficient data representation. They show where design decisions are complicated (coupled) and where they are not. The problem of developing a set of uncoupled or decoupled design matrices spanning the design space for our problem is as complex as solving the design problem by using structured design.

Following the principle of structured or axiomatic design, one could easily see a phased plan, perfectly fit for a work breakdown structure and presumably fit for effective and efficient development process. Industrial practice shows that this approach often results either in risk-adverse incremental development of a known concept, or in cyclic hard to finalize development processes in case a new concept was pursued. It is very hard to predict up-front what dependencies in which concepts are vital to a successful design. As a consequence, the product of architectures of cars and aircraft have not changed significantly over decades of their existence. Despite claims that technology developments are speeding up, the impact of new technology or new materials is often limited to redesign of subsystems. The problem of introducing new and unknown relations are avoided as much as possible. The opportunities of new business models are the scope of implementing new technologies into existing product platforms.

 4.2 Concepts of Self Organization

In the area of systems control and cybernetics, self organization refers to systems that are capable of changing their structure and their functionality on order to adapt to new environments.  Another perspective on self organization originates from a systems perspective on understanding nature, life and organizations. This perspective, called autopoiesis (self-production), does not take adaptation as a response to changes in the environment as axiom, but it claims that living structures influence or adapt to their environment as a means to self-maintain and improve their chance of reproduction.


An autopoietic machine is a machine organized (defined as a unity) as a network of processes of production (transformation and destruction) of components that produce components which:

  1. through their actions and transformations continuously regenerate and realize the network of processes (relations) that produced them; and
  2. constitute it (the machine) as a concrete unity in the space in which they (the components) exist by specifying the topological domain of its realization as such a network.

Evolutionary Problem Solving

Evolutionary problem solving is based on the structure of genetic algorithms. The basic structure of the genetic algorithm, originally developed by Holland is as follows:

  1. Initialize a starting population of physically feasible solutions.
  2. Create a new generation through genetic operands such as mutation, crossover and reproduction
  3. Rank this population using the fitness function.
  4. Select the top of this population and randomly select a couple of others to create a new starting population
  5. Repeat steps 2-4 until top-member of a generation has sufficient fitness score to be acceptable as a solution.

The success of this nature-inspired algorithm can be attributed to two properties that make it distinct from linear optimization techniques. The first property is redundancy and diversity. Rather than developing one solution, genetic algorithms develop and maintain multiple solutions concurrently. The resulting diversity is needed to maximize the probability to have solutions available at all times that can comply with all requirements and constraints. The second property is the non-linearity of the selection process. With linear optimization the fitness landscape is set from the start by the starting solution and the fitness function. Finding the optimum in this landscape could mean an exhaustive search through the entire landscape. Although the fitness landscape is set from the start, a genetic algorithm employs multiple starting points for the search, and the generation and selection steps cause the effective fitness landscape to be reshaped at the start of each generation.

Evolutionary Problem Solving and Self Organization

The genetic algorithm (GA) is a system model for the self-reproduction principles of the autopoiesis theorem. The solution patterns a GA may generate are predominantly the result of the initial set of solutions that were present at the start. The fitness function is the context within which structure changes may occur as long as survival as a unity or species is not at stake. Changing the fitness function will cause serious changes and may also cause death in case the present elements can not generate a sufficient fit (survival) to the new fitness function. in case of survival, biologists may recognize evolution, but they may also claim that the new organism is a different unity that is capable of a different set of interactions, fit for the new context. Thus the old species is declared extinct since it evolved into a new distinguishable organism.

4.3 Set-based Concurrent Engineering

Ward and his co-authors argue that in concurrent engineering there are two fundamentally different approaches to be recognized: point-based and set-based. In case of point-based design, a single solution is synthesized first, then analyzed and changed accordingly.  Even though the phases of the design process may be executed concurrently, all designers and specialists are investing their efforts in the pursuit of only one concept that is to be developed into a solution.

In set-based concurrent engineering, designers explicitly communicate and think about sets of design alternatives at both conceptual and parametric levels. The efficiency of set-based versus point-based design is that in communicating sets, implicitly or explicitly, all designers become more focused on relations and constraints between different aspects of the design than they would be when focusing at a point solution. All designers communicate their range of options rather than one preferred option. Sometimes to maintain focus, constraints for these sets can be set tighter than they would be in case of a point based design.

Set-based Concurrent Engineering: Toyota and Nippondenso

The set-based engineering process of Toyota have the following characteristics:

  1. The team defines a set of solutions at the system level rather than a single solution
  2. It defines sets of possible solutions for various subsystems.
  3. It explores these possible solutions in parallel, using analysis, design rules and experiments to characterize a set of possible solutions.
  4. It uses analysis to gradually narrow the sets of solutions. In particular the team uses analysis of the set of possibilities for subsystems to determine appropriate specifications to impost on those subsystems
  5. Once the team establishes a single solution for any part of the design, it does not change it unless absolutely necessary


Nippondenso. . . .also applies a process that has characteristics of set-based concurrent engineering and extends this even to pre-design R&D. In this process, the degree of parallelism and redundancy is much higher than it typically is with Toyota.

As an automotive supplier, the demand for diversity is higher and their competitiveness is much affected by new technologies and new materials. In order to push the limits and to stay ahead of the competition, Nippondenso tests as many ideas as they can to create a platform (set) of solutions that is competitive and can be easily adapted to the specific interfacing requirements of different car makes. What may be a surprise is that the start of Nippondenso’s development processes may be 3-5 years ahead of the start of the car development processes that adopt the new designs. Rather than pursuing rapid development once the outline of specification from their customers is clear, Nippondenso pursues radical breakthrough designs that are ready before their customers ask for them. When they start working with their customers, the focus is on interfacing and not on the core technology, which enables them to avoid the major part of development risk.

4.4 Evolutionary Organization of Design Process

The process can be as follows:

  1. Divide the staff into n independent teams that are all capable of executing the entire project, and give all these teams an identical assignment and a deadline.
  2. The teams will develop their concepts and solutions following set-based concurrent engineering practices, and they will record their achievements and findings in lessons learned books.
  3. At regular intervals, a fair is organized where all teams present their progress and give insight int their lessons learned books.
  4. At these fairs, team members look around for promising partial solutions with their “competitors”
  5. After the fair, teams continue their own development, including ideas inspired by the last fair.
  6. If a design with sufficient fitness has been achieved, stop; else, repeat steps 3-5.

The processes within the teams could also have the characteristics of a genetic algorithm if they apply brainstorming for finding and selecting ideas. However, the fair is really the place where crossovers and mutations occur. At the fair, everyone is looking for clever ideas that could fit to their own concept (crossings), and some ideas may also trigger new thoughts (mutations). Although a fitness function that could be used for ranking exists, the organizational form of a GA has the advantage that the ranking of partial ideas is fuzzy and not explicit. This means that ideas that may not be very successful in one context could be a perfect fit in another context. In case of explicit ranking, these ideas could have been lost. The implicit ranking also solves a social problem of working with a large engineering group force, where a dozen socially dominant engineers will monopolize the decision making at centralized meetings to a degree where a significant portion of the engineering staff effectively has no influence. Because the central meeting is now a fair where implicit recognition is the mechanism for the survival of ideas, good ideas, regardless of their source, stand a good chance of being inherited into the final concept. The process can be made more efficient if overlaid with a structured design process where the progress at the exchange moments (fairs) becomes synchronized.

Leave a Reply