Joint Research Conference
June 24-26, 2014
Over the last decade, the production industry has experienced a change from a supply-oriented to a demand-oriented design of products. As a result, an efficient adaptation of the available prod- ucts and processes to the changing customer needs is an important requirement for industrial success. Therefore, a multi-objective optimization with respect to conflicting workpiece proper- ties, e. g., hardness and ductility, can increase the flexibility with respect to the achievable prop- erty distributions and combinations. Until now, only a few very basic approaches for sequentially planning experiments in a multi-objective context are available. More complex techniques do ex- ist, but are usually limited to two objectives and/or expensive to compute. The proposal of criteria that are both, efficient to evaluate and scalable with the number of objectives, is one important contribution of this talk. Two new criteria for sequentially choosing experimental designs for the approximation of the Pareto set are developed. They dispense with a tedious integration for a quick evaluation and improved scalability with the number of objective dimensions. Based on a list of necessary conditions and required properties, the new and some established infill criteria can be formally evaluated. A comprehensive benchmark is conducted in order to assess the perfor- mance when applied in a sequential optimization.