Investigating the Cognitive Behavior of Generating Idea Sketches. Neural Network Simulation

In idea sketches, there are a number of ambiguous shapes. Designers will associate and transform some shapes into others (Liu, 1993). Then, they evaluate these shapes in terms of functions and design requirements, furthermore, they would have generated other shapes that certified the design requirements (Huang, 1999). However, not only is the idea of design composed of one element, but also consisted of varied components. The purpose of this paper is to investigate how designers generate ideas of multi-component products, and to simulate this phenomenon by neural networks. At the same time, this paper attempts to study the design cognitive behaviour of idea-generating stages, and explores the designers'cognitive phenomenon. Therefore, there are two stages in this paper: First, I conduct a cognitive experiment to realize how designers generate the multi-component product and acquire the sketches that designers generated. Second, I train the neural networks to simulate the behaviour of idea generation and explore the cognitive phenomenon in design sketches. As a result, networks associate one shape that trained before, and then generate a complete idea. This phenomenon is similar to the cognitive behaviour of designers who saw the ambiguous shape as one shape, which was retrieved from LTM. Moreover, the neural network is examined by a rectangle, which is totally different from the training patterns. The network will associate a confused shape. But the network will associate different shapes by adjusting some critical parameters. Designers can generate variable shapes from one shape, but the signal neural network can't simulate this kind of behaviour. On the contrary, this paper proposes five sequential networks to generate variable shapes from the same shape and simulates how designers develop ideas.