Multiobjective Metamodel-Based Design Optimization—A Review and Classification Approach Using the Example of Engine Development

Abstract

<div class="section abstract"><div class="htmlview paragraph">To cope with increasing, challenging requirements and shorter development cycles, more complex, often nonlinear, systems with high interactions have to be optimized in many fields of research, such as the energy sector. As this often goes beyond the classical parameter studies-based approach, systematic optimization approaches offer a key solution. In the context of the development of energy converters, like engines, such techniques are applied to enhance efficiency and enable optimal use of energy. This review provides a comprehensive overview of the field of optimization approaches, more precisely referred to as Metamodel-Based Design Optimization (MBDO). The MBDO approaches essentially comprise three main modules: the Design of Experiment (DoE), the Response Surface Modeling (RSM), and the Multiobjective Optimization (MoO), in varying compositions. Previous reviews primarily focused on a selection of these modules, whereas this novel review equally covers and structures the modules DoE, RSM, and MoO and their combination to MBDO approaches. Many examples of these modules and MBDO implementations and their interrelationship, strengths, and limitations are discussed in detail and supplemented with many exemplary methods, e.g., from engine development. Methods from previous reviews are collected and updated with recent approaches, e.g., including new machine learning methods used in this context. Moreover, this study presents a holistic, extended classification approach to structure any MBDO method. The classification, which is based on the existence, structure, and interactions of the modules DoE, RSM, and MoO, is applied to various MBDO approaches from the literature. One recent MBDO focus of research is the development of online adaptive approaches as these allow to use valuable information obtained during the optimization process to guide the DoE or MoO. Therefore, the online adaptivity, feedback loops, and strengths and limitations of MBDO approaches are a novel focus area of this review. Recommendations and requirements for future “Fully Online MBDO” approaches with enhanced adaptability and generalizability are derived.</div></div>