Background


  • Laboratory of Scientific Computing: strong traditions in FEM, optimal control, PDEs and modelling.
  • Need of knowledge in optimization in solving real-world applications (simulation was not enough), in particular, nonsmooth and multiobjective cases
  • First separately, then combined.
  • New methods should be: computationally efficient, interactive, easy to use etc.

Method development - Subgradient GDF


  • Theoretically a very straightforward approach.
  • Combination of the Geoffrion-Dyer-Feinberg method and the subgradient method.
  • Publications: Optimization Methods and Software (1993), Proceedings of ECMI (1994).
  • Applications: deflection of a string, continuous casting process.

Method development - NIMBUS version 1


  • Emphasis on easiness of use, not theoretical soundness. What can we expect DMs to be able to say?
  • A natural extension of the proximal bundle method -> MPB (constraints and objective functions treated similarly, scalarization hidden).
  • Classification into 5 classes.
  • Feasible region assumed to be convex.
  • Publications: Optimization (1995), Proceedings of IFIP (1996), MOPGP94 (1996), MCDM94 (1997)
  • Applications: problems from the literature, thickness of a beam.

Method development - NIMBUS version 2


  • The same classification idea but a scalar subproblem.
  • Feasible region assumed to be compact.
  • The current interactive WWW-NIMBUS on the Internet http://nimbus.mit.jyu.fi
  • Publications
    • Proceedings of MCDM97 (1998): WWW-NIMBUS
    • Computational Optimization and Applications (1998): application: continuous casting process
    • JORS (1999): comparison of the two NIMBUS versions and the Reference Direction method, application: 48 problems from the literature
    • Computers and Operations Research (2000): WWW-NIMBUS including a technical description

Activities and projects in the past


  • Partner of the European Research Project DECISION (1997-2000)
    • ESPRIT Research Program: new robust optimizers for industrial design purposes in heterogeneous computing environments
    • Partners: NAG Ltd., INRIA, VTT, Nokka-Tume, EMFiTech and Dassault Aviation
    • Applications: optimal shape design of grapple loaders, flat EMFi-actuators, wing profile of an aircraft
  • Node of the European Thematic Network INGENET (1997-2001)
    • Brite Euram Programme: networked industrial design and control applications using genetic algorithms and evolution strategies
    • 22 European partners (industry & academy)
    • Applications - database of test cases: optimal shape design of a headbox of a paper machine, aerodynamic shape optimization, pattern recognition techniques etc.
  • Organizer of a Short Course on Evolutionary Algorithms in Engineering, EUROGEN99
  • Multiple Criteria Decision Making in Environmental Planning and Management Problems
  • Multiobjective optimal control of the cooling process of continuous casting of steel

Related lectures


  • Course on optimality conditions, nonsmooth and multiobjective optimization and genetic algoritghms with applications, that is, kevään 2000 OPTIMOINNIN JATKOKURSSI
  • Course on Nonlinear Programming, that is, syksyn 1999 OPTIMOINTI-kurssi