Session: DAC-13-1: Metamodel-Based Design Optimization (MBDO)

Paper Number: 141932

141932 - Data-Driven Sizing and Shaping of Topology Optimization Concepts Using Implicit Surfaces, Free Form Deformations and Multifidelity-Based Surrogate Models 

In this work, a framework for data-driven sizing and shaping of topology optimization (TO) concepts is developed, implemented and demonstrated. The density field from a solid isotropic material with penalization (SIMP)-based TO solution is converted to an implicit surface-based geometry (ISG) by using regularized radial basis function networks (RBFN) with Wendland's compactly supported radial basis functions. Sizing of the ISG is done locally by morphing operations and shaping is performed by applying free form deformations (FFD) on the stl-mesh which is generated from the ISG by a marching cube algorithm. The smooth FFD-based shaping is represented as a RBFN with cubic splines using a set of control points with corresponding prescribed deformations. Data-driven sizing and shaping of TO concepts are then performed by using multifidelity non-linear computer experiments and surrogate model-based design optimization. The developed and implemented framework is demonstrated for the well-known Messerschmitt-Bölkow-Blohm (MBB)-beam as well as an application of a flywheel to a compactor machine.

Topology optimization performs best for linear state models and few design constraints. Therefore, multiple design requirements needing non-linear state models are usually excluded in the conceptual phase using topology optimization. Instead the TO concept is adjusted afterwards to satisfy these additional design requirements using non-linear state models in a following detailed design phase. This work is often tedious using trial-and-error strategies, and the new design result might be far locking from the original TO concept. Typically, the standard parametric CAD model is one of the main reasons for this, because of the difficulties of representing the TO results properly in a standard CAD environment. In this work a data-driven framework is suggested to overcome these issues by converting the TO design result to implicit surface-based geometries and then perform detailed size and shape optimization by using morphing, applying free form deformations and training multifidelity surrogate models to be used in a detailed design optimization formulation replacing the tedious trial-and-error strategies discussed above.

Presenting Author: Niclas Strömberg Örebro University

Presenting Author Biography: Niclas Strömberg is a professor in Engineering mechanics. His research is focused on optimization driven design using finite element analysis, topology optimization and metamodel-based design optimization.

Authors:

Niclas Strömberg Örebro University