A metamodeling approach towards virtual production intelligence

Decision making for competitive production in high-wage countries is getting strongly influenced by computer simulations in order to cope with the fast changing global market demands. These simulations offer the possibility to find better designs within a short time, predict and optimize properties that cannot be measured directly from experiments, and explore new information and parameter ranges in known processes that are not easily accessible experimentally. Virtual prototyping is being successfully introduced in production where new products are virtually designed on the computer before building expensive machine components. This led to an explosion of tools that are being unified into a new notion in production, the virtual production system. This virtual system is based on the unification of real (experiments) and virtual (simulations) worlds, bringing together the scientific knowledge and technical know-how about the process. The outcome is presented by operative tools, namely process maps and visual design apps targeted to get more skilled developers and operators. The primary achievement of this thesis is associated with the generation of process maps and visual design apps that are capable to explore the design space spanned by solutions of complex simulation models with the aid of a fast and frugal model, known as a metamodel. Metamodeling follows Benjamin Franklin’s rule for decision making formulated in 1772 [46], where he defined "Prudential Algebra", to be one of the major ingredients that can be applied in order to obtain one algebraic Pareto-value labelling alternative decisions. Metamodeling techniques define a procedure to analyze and represent complex physical systems using easy to use, fast mathematical designs to create cheap numeric surrogates that describe cause-effect relationships between setting parameters as inputs and criteria like product quality variables as outputs. This dissertation focuses on two main research questions and three achievements. The first research question focuses on how to efficiently generate and validate the approximation quality of a metamodel representing a complex simulation model. The question is investigated by different sampling, interpolation, and validation techniques. The outcome is a novel iterative method that generates a high quality metamodel for outputs with continuous as well as piecewise responsesthe first achievement. It basically combines the Sequential Approximate Optimization (SAO) procedure and the Radial Basis Function network (RBFN). The second research question focuses on how to improve a complex model by replacing the simulation model by a metamodel. This question is investigated by interfacing the RBFN metamodel with several global sensitivity analysis and visualization techniques. The outcome is a "Design-Cockpit" used for Virtual Production Intelligencethe second achievement. Its main advantage is exploring information from complex multidimensional computer simulation models for optimizing the process and improving the know-how as well as the model structure with respect to the relevant properties. Finally, the concept of process maps and visual designs are applied to five different laser manufacturing applications with different parameter domain spaces ranging from two to seven dimensions of parameter space. The results show a great efficiency of process map and a visual design in supporting decision makingthe third achievement. To the author’s best knowledge, no fast interactive process map is conducted in any publication of laser manufacturing process before. Although the process maps and underlying process apps are applied to laser applications, the metamodeling techniques can be applied to almost any economical, ecological, or technical process, where the process itself is described by a black box model giving scalar data.