• Computational Intelligent Design

Computational Intelligent Design

 

Decisions in design and engineering are difficult to take due to complexity that generally arises from the following three issues. 

  • The first issue is the involvement of multiple decision criteria, such as operational certainty, financial certainty, sustainability, or attractiveness, some of which are conflicting and some have a soft character. That is, the relation between the variables subject to decision making and the objectives at hand may be nonlinear and involve uncertainty. These issues make computational treatment challenging.
  • The second issues is the involvment of multiple, stiff constraints that must be satisfied, such as time, money and space restrictions. The stiffness refers to large numerical difference among several constraints subject to minimization. Satisfying multiple stiff constraints at the same time, while also satisfying the objective at hand is very difficult to accomplish.
  • And finally, the third source of complexity is the involvement of several independent variables constituting a solution, which implies an excessive amount of possible solutions. In combination with the two issues mentioned above, this imposes a severe challenge for computational methodologies, to yield solutions in a fast, effective, and robust manner.

These issues make it formidably challenging to reach most suitable solutions. This diffculty is alleviated when advanced computational methods are used to deal with the complexity. This is the subject matter of the computational intellgent design research presented here. In particular methods from the domain of computational intelligence, such as evolutionary and neural computation, are employed to deal with soft and conflicting objectives, stiff constraints and vast solution domains. As result, solutions are guaranteed to satisfy the objectives at hand as much as possible, while they satify the constraints at the same time. This quality assurance is highly desirable in the face of depleating resources and increasing demands imposed on engineering and design products, and it will become more and more relevant in the future, in proportion with the increase in complexity of the real-world decision-making problems.

Recent peer-reviewed international publications (2006-2012)

Book chapters

  1. Bittermann, M.S., Sariyildiz, I.S., Ciftcioglu, Ö.: A computational intelligence approach to alleviate complexity issues in design. Portugali, J. and Meyer, H. (eds.) in: Complexity Theories of Cities have come of Age - Part Two: Implications to Planning and Urban Design. Springer Verlag, Heidelberg (2012)
  2. Ciftcioglu, Ö., Bittermann, M.S.: Adaptive formation of Pareto front in evolutionary multi-objective optimization. In: Lazinica, A. (ed.): Evolutionary Computation. In-Tech, Vienna (2009)
  3. Ciftcioglu Ö., Bittermann M.S.: From perceptual towards cognitive robotics in the framework of evolutionary computation. In: Pennacchio, S. (ed.): Emerging Technologies, Robotics and Control Systems. InternationalSAR, Palermo, Italy (2009)
  4. Ciftcioglu, O.: Multiresolutional Filter Application for Spatial Information Fusion in Robot Navigation. Robotics, Automation and Control. IN-TECH Publishing, Vienna (2008)
  5. Bittermann, M.S., Sariyildiz, I.S., and Ciftcioglu, Ö.: Blur in Human Vision and Increased Visual Realism in Virtual Environments. In: Lecture Notes on Computer Science: Springer Verlag (2007)
  6. Ciftcioglu, Ö.: Shaping the Perceptual Robot Vision and Multiresolutional Kalman Filtering Implementation. In: Emerging Technologies, Robotics and Control Systems International Society for Advanced Research (2007)
  7. Ciftcioglu, Ö., Bittermann, M.S., and Sariyildiz, I.S.: Visual perception theory underlying perceptual navigation. In: Emerging Technologies, Robotics and Control Systems International Society for Advanced Research (2007) 139-153
  8. Sariyildiz, I.S., Bittermann, M.S., and Ciftcioglu, O.: Perception & Architecture. In: The Architecture Annual 2005-2006 Delft University of Technology, H. Bekkering, D. Hauptmann, A. d. Heijer, J. Klatte, U. Knaack, and S. v. Manen, Eds. Rotterdam: 010 Publishers (2007) 104-109
  9. Ciftcioglu Ö and Sariyildiz I.S Fuzzy logic for stochastic modeling in Soft Methods for Integrated Uncertainty Modelling, Advances in Soft Computing, J. Lawry, E. Miranda, A. Bugarin, S. Li, M. A. Gil, P. Grzegorzewski, and O. Hryniewicz, Eds. Tokyo: Springer, 2006.

Journal papers

  1. Datta, R., Bittermann, M.S., Deb, K., Ciftcioglu, Ö.: Probabilistic Evolutionary-Classical Constraint Optimization with Robotics Application. Robotics and Computer-Integrated Manufacturing (2012) (to be published)
  2. Ciftcioglu, Ö., Bittermann, M.S.: From perceptual towards cognitive robotics in the framework of evolutionary computation. Int. J Factory Automation, Robotics and Soft Computing (2009)
  3. Bittermann M.S., Ciftcioglu Ö.: Visual perception model for architectural design. Journal of Design Research 7 (2008) 35-60
  4. Ciftcioglu, Ö.: Shaping the Perceptual Robot Vision and Multiresolutional Kalman Filtering Implementation. Int. Journal Factory Automation, Robotics and Soft Computing (2008)
  5. Bittermann M.S., Sariyildiz I.S. and Ciftcioglu Ö. Visual perception in design and robotics. J Integrated Computer-Aided Engineering, 14 (2007) 73-91
  6. Ciftcioglu, Ö., Bittermann, M.S., and Sariyildiz, I.S.: Visual perception theory underlying perceptual navigation. Int. J. Factory Autom., Robotics and Soft Comp. (2007) 171-185
  7. Ciftcioglu Ö.,  Bittermann M.S. and Sariyildiz I.S. Multiresolutional fusion of perceptions for perceptual robotics. Journal of Advanced Computational Intelligence and Intelligent Informatics (JACIII) 11 (2007) 688-700

Conference papers 

  1. Ciftcioglu, Ö., Bittermann, M.S., and Sariyildiz, I.S.: A probabilistic approch for constrant optimization by genetic algrithm. Proc of Unconventional Computation & Natural Computation, Oreans, France, September 3-7 2012,  Lecture Notes in Coputer Science, LNCS, Spinger (2012) (to be pulished)
  2. Datta, R., Bittermann, M.S., Deb, K., Ciftcioglu, Ö.: Probabilistic Constraint Handling in the Framework of Joint Evolutionary-Classical Optimization with Robotics Applications. IEEE Congress on Evolutionary Computation at World Congress on Computational Intelligence – WCCI, the IEEE, Brisbane, Australia, June 10-15 (2012) (to be published)
  3. Bittermann, M.S., Ciftcioglu, O.: A cognitive system based on fuzzy information processing and multi-objective evolutionary algorithm. IEEE Conference on Evolutionary Computation - CEC 2009. IEEE, Trondheim, Norway (2009)
  4. Ciftcioglu Ö., Bittermann M.S.: Solution Diversity in Multi-Objective Optimization: A study in Virtual Reality. World Congress on Computational Intelligence WCCI 2008, Hong Kong (2008)
  5. Ciftcioglu Ö.: A fuzzy neural tree for possibilistic reliability. Joint 4th Int. Conf. on Soft Computing and Intelligent Systems (SCIS & ISIS), Nagoya, Japan (2008)
  6. Ciftcioglu Ö., Bittermann M.S.: Multi-objective optimization for cognitive design. Joint 4th Int. Conf. on Soft Computing and Intelligent Systems (SCIS & ISIS), Nagoya, Japan (2008)
  7. Sariyildiz I.S., Bittermann M.S., Ciftcioglu Ö.: Multi-objective optimization in the construction industry. AEC 2008, Antalya, Turkey (2008)
  8. Sariyildiz, I.S., Bittermann, M.S., Ciftcioglu, Ö.: Performance-based Pareto optimal design. In: Rusák, I.H.a.Z. (ed.): TMCE 2008, Izmir, Turkey (2008)
  9. Ciftcioglu, Ö., Bittermann, M.S., and Sariyildiz, I.S. Building performance analysis supported by GA. Proc. 2007 IEEE Congress on Evolutionary Computation, Singapore (2007) 489-495
  10. Ciftcioglu, Ö. and Sariyildiz, I.S.: Further studies on visual perception for perceptual robotics. Proc. Fourth Int. Conf. Informatics in Control, Automation and Robotics - ICINCO2007, Angers, France (2007) 468-744
  11. Ciftcioglu, Ö., Bittermann, M.S., and Sariyildiz, I.S.: Sensor data fusion in autonomous robotics. Proc. The 2nd Int. Conf. Innov. Comp., Inf. and Contr. - ICICIC 2007, Kumamoto, Japan (2007)
  12. Ciftcioglu, Ö., Bittermann, M.S., and Sariyildiz, I.S.: Fuzzy neural tree for knowledge driven design. Proc. The 2nd Int. Conf. Innov. Comp., Inf. and Contr. - ICICIC 2007, Kumamoto, Japan (2007)
  13. Ciftcioglu, Ö., Bittermann, M.S., and Sariyildiz, I.S.: A neural fuzzy system for soft computing. Proc. NAFIPS 2007, San Diego, USA (2007) 489-495
  14. Ciftcioglu Ö, Bittermann M.S and Sariyildiz I.S Towards computer-based perception by modeling visual perception: a probabilistic theory. Proc. IEEE International Conference on Systems, Man and Cybernetics, October 8-11, 2006, Taipei, Taiwan.
  15. Ciftcioglu Ö, Bittermann M.S and Sariyildiz I.S Fusion of perceptions for perceptual robotics. Proc. NAFIPS’06, June 3-6, 2006, Montréal, Québec, Canada.
  16. Ciftcioglu Ö, Bittermann M.S, and Sariyildiz I.S Studies on visual perception for perceptual robotics. Proc. ICINCO 2006, 3rd Int. Conference on Informatics in Control, Automation and Robotics, August 1-5, 2006, Setubal, Portugal.
  17. Bittermann M.S and Ciftcioglu Ö Real-time measurement of perceptual qualities in conceptual design. Proc. 6-th Internationl Symposium on Tools and Methods of Competitive Engineering TMCE 2006, April 18-22, 2006, Ljubljana, Slovenia.
  18. Bittermann M.S, Sariyildiz I.S, and Ciftcioglu Ö Visual space perception model Identification by evolutionary search. Proc. 9-th International Design Conference - Design 2006, May 15-18, 2006, Dubrovnik, Croatia.
  19. Ciftcioglu Ö, Bittermann M.S and Sariyildiz I.S Application of a visual perception model in virtual reality (poster). Proc. ACM SIGGRAPH Symposium on Applied Perception in Graphics and Visualization, APGV’2006, July 28-30, Boston, USA, 2006.
  20. Bittermann M.S and Ciftcioglu Ö. Validation of a visual perception model. Proc. Joint International Conference on Construction Culture, Innovation, and Management (CCIM), 26-29 November 2006, Dubai, United Arabian Emirates.
  21. Ciftcioglu Ö and I. Sevil Sariyildiz Knowlegde model for knowledge managememnt in the construction industry. Proc. Joint International Conference on Construction Culture, Innovation, and Management (CCIM), 26-29 November 2006, Dubai, United Arabian Emirates.
  22. Ciftcioglu Ö and Sariyildiz S.I On the efficiency of multivariable TS fuzzy modeling. Proc. 2006 IEEE International Conference on Fuzzy Systems, July 16-21, 2006, Vancouver, BC, Canada.
  23. Ciftcioglu Ö and Sariyildiz S.I On the efficiency of fuzzy logic for stochastic modeling. Proc. NAFIPS’06, June 3-6, 2006, Montréal, Québec, Canada.
  24. Ciftcioglu Ö, Bittermann M.S and Sariyildiz I.S Autonomous robotics by perception. Proc. ISCIS & ISIS 2006, Joint 3rd International Conference on Soft Computing and Intelligent Systems and 7th International Symposium on Advanced Intelligent Systems, September 20-24, 2006, Tokyo, Japan.
  25. Ciftcioglu Ö, Bittermann M.S and Sariyildiz I.S Fuzzy ARX modeling of dynamic systems. Proc. ISCIS & ISIS 2006, Joint 3rd International Conference on Soft Computing and Intelligent Systems and 7th International Symposium on Advanced Intelligent Systems, September 20-24, 2006, Tokyo, Japan.

Multi-dimensional decision analysis

Evaluation of the a decision needs consideration of many facts at the same time. A multi-dimensional analysis model is a model to compute the suitability of a decision, where every dimension refers to a certain decision aspect. Using this model the suitability of a decision is computed with respect to multiple dimensions at the same time. A method to establish such a model is a neuro fuzzy modeling.

      

Multi-dimensional analysis model using a neuro-fuzzy systemknown as fuzzy neural tree    

      

Nonlinear mapping at an inner node of a fuzzy neural tree    

A neural fuzzy system for soft computing. Proc. NAFIPS 2007, San Diego, USA (2007) 489-495 

In a neuro fuzzy system the linguistic concepts involved in the decision analysis, such as sustainability, functionality, etc. are represented by means of neurons performing a non-linear mapping from the neuron's input to its output, simulating a reasoning activity in a brain.The neuro fuzzy system we developed for decision analysis is different from artificial neural networks in the sense that the latter are established using training data to optimize the model parameters being the weight connections among nodes, whereas the method we developed optimizes the activation function in the neuron while the connections weights are fixed. This way our method is able to model deep knowledge from already existing knowledge, whereas ANN identify knoweldge from data, i.e. ANN are used for data-driven knowedge modelling, while fuzzy neural trees are used for knoweldge-driven knowledge modelling.

Illustration of fuzzy information processing at a neuron to obtain the design performance.

A cognitive system based on fuzzy information processing and multi-objective evolutionary algorithm , IEEE Congress on Evolutionary Computation –CEC 2009, Trondheim, Norway, 18-21st May 2009

Computational Intelligence for Enhanced Decision Making in Engineering and Design

In this approach a multi-dimensional performance model is integrated into a computational search process.

Multi-dimensional performance-based design by means of a cognitive system using evolutionary computation & fuzzy logic

A cognitive system based on fuzzy information processing and multi-objective evolutionary algorithm, IEEE Congress on Evolutionary Computation –CEC 2009, Trondheim, Norway, 18-21st May 2009

This way design parameters are incorporated into a complex algorithm, namely evolutionary algorithm with fuzzy neural computation, that finds the best set of solutions to meet the objectives set by the design team.

Visual representation of optimal solutions for two objectives in the urban design

The solutions obtained in this way are known as Pareto-optimal front. They provide a variety of outstanding alternatives to a decision maker, since none of these solutions is outperformed by another one. Every solution is equally valid, and a decision maker selects among them with great confidence.

Visual representation of optimal solutions for four objectives in the interior design task

As a computational solution set has been built, alternate designs are explored by varying the parameters. The generative system can handle effectively up to five objectives, and has no restriction regarding the number of variables playing role on the objectives. The amount of variables characterizing a solution is only limited by available computational time and power. For more than four objectives the five-dimensional Pareto front can be represented as well.

Visual representation of optimal solutions for five objectives

The strength of the approach is that solutions can be assessed without any presupposition, and confidence of finding the best solution is increased. Human and computational cognitive system are in an interaction loop: Human decision maker is setting the criteria, computations privide optimal solutions for these, based on these solutions the decision maker modifies criteria and so on, until a Pareto-optimal solution matches the designer's complete preferences as far as possible.

Two Pareto optimal solutions generated by a multi-dimensional performance-based design system

Solution Diversity in Multi-Objective Optimization: A study in Virtual Reality. World Congress on Computational Intelligence WCCI 2008, Hong Kong (2008)

One of the Pareto optimal solutions for an interior design task, where visual perception plays a role in the computational optimisation 

A cognitive system based on fuzzy information processing and multi-objective evolutionary algorithm, IEEE Congress on Evolutionary Computation –CEC 2009, Trondheim, Norway, 18-21st May 2009

Perception modeling

Architectural design involves perception-based requirements, such as visual openness or visual privacy. Such requirements are challenging to treat, because the human vision process is highly complex, involving brain processes. Therefore the comparison of perceptual properties among scenes is imprecise.

To let perception play a more prominent role in design, a model of human vision is developed. The model is based on probabilistic terms. This way the complexity of the vision process is absorbed.

Unbiased visual attention for a nearby object

Unbiased visual attention for a distant object

The model is implemented by means of an avatar in virtual reality. The avatar experiences the environment in a human-like manner, so that the results are used during the evaluation of design alternatives.

From perceptual towards cognitive robotics in the framework of evolutionary computation. In: Pennacchio, S. (ed.): Emerging Technologies, Robotics and Control Systems. InternationalSAR, Palermo, Italy (2009)

Probability density of perception for objects that are oriented perpendicular to the observer's forward direction

Visual perception theory underlying perceptual navigation. In: Emerging Technologies, Robotics and Control Systems International Society for Advanced Research (2007) 139-153

Perception measurement in virtual reality.

Towards computer-based perception by modeling visual perception: a probabilistic theory. Proc. IEEE International Conference on Systems, Man and Cybernetics, October 8-11, 2006, Taipei, Taiwan.

Perception analysis of an interior space

The authors are with the Chair of Design Informatics, Delft University of Technology, Faculty of Architecture, Dept. of Architectural Engineering and Technology,Julianalaan 134, 2628 BL Delft , NL  | Email:i.s.sariyildiz@tudelft.nl ;m.s.bittermann@tudelft.nl ;o.ciftcioglu@tudelft.nl   | Tel: +31-6 45728249

© 2012 TU Delft

Metamenu