A COMPARATIVE STUDY: STRUCTURAL COMPLEXITY METRICS APPLIED AGAINST Function and Assembly PRODUCT GRAPHS in Predicting Market Price and Assembly Time

The overarching objective of this research is to investigate the measurement and use of complexity in the prediction of product performance metrics (assembly time and market cost) for two model graph types (assembly models and function structures). This research focusses on analyzing how accurate the prediction of performance metrics are based on these graph types. This research focuses on developing four prediction models: Function Structures to predict Market Price (FS-MP), Assembly Models to predict Assembly Time (AM-AT), Function Structures to predict Assembly Time (FS-AT), and Assembly Models to predict Market Price (AM-MP). These assembly models and function structures are analyzed against twenty-nine complexity metrics resulting in a complexity vector, which in turn, is used to train a population of 18,900 artificial neural networks (ANN). The ANNs serve as surrogate models to map these graphs to performance values. The models are created with a common database of products that are readily available in the market, such as consumer electro-mechanical products, power tools, kitchen appliances, or children\u27s toys. The overarching goal of this research is to assist designers in product development by providing information earlier in the design process that is not currently available. For example, in early design stage when engineers are developing different functional concepts, it is currently impossible to compare them based on cost. However, with the graph based historically trained complexity approach, one can compare different function structures in terms of market price. It is also not known whether the function structures could also be used to predict assembly time, a major contributor to manufacturing cost. Furthermore, the assembly models of the selected concepts are created in the embodiment design stage which can be used in the accurate predictions of the performance metrics (assembly time and market value). These models are based on more information and understanding of the design problem, and should therefore result in more accurate predictions of the performance metrics than those resulting from the conceptual design stage information. Ultimately, based on the understanding of how accuracy in prediction models change based on graph input type and on performance type, one could envision generating multiple different historically based predictors that can inform design earlier in the process