Data-Driven and Real-Time Prediction Models for Highly Iterative Product Development Processes

Abstract

Some high-level technical products are associated with transdisciplinary simulation-driven design processes. Therefore, their design process involves many stakeholders and is prone to frequent changes, leading to a highly iterative process with a long lead time. Despite the decades of statistical approximations and metamodeling techniques on prediction models, companies are still striving to achieve fully automated real-time predictions in early design phases. The literature study shows a gap in existing methods such as not being fully real-time or suffering from high dimensionality. This paper presents a generic model for the development process of such described products and motivation for such modeling through a series of semi-structured interviews with an automotive sub-supplier company. The proposed process model points to the digital verification in every design loop as the bottleneck which is then confirmed by interviewees. As alternative solutions to overcome the problems, a method for data-driven and real-time prediction models is presented to enable the designer to foresee the consequence of their decision in the design phase. To evaluate the method, two examples of such real-time meta-modeling techniques, developed in an ongoing research project are discussed. The proposed examples confirm that the framework can reduce lead time spent on digital verification and therefore accelerate the design process in such products.