Research on Human-Machine Collaborative Aesthetic Decision-Making and Evaluation Methods in Automotive Body Design: Based on DCGAN and ANN Models

The main content of this study is the human-machine collaborative design research, taking the car body design as the carrier. The research framework focused on two phases of car body design process, that of design ideation and evaluation. In the ideation stage, we trained an imperfect Deep Convolutional Generative Adversarial Network (DCGAN) model that just could generate blur automobile images as the blur design motherboards for the iterative sketching, which had design uncertainties and blanks, thus activating designers’ subjective initiative and aesthetic intuition to provide more creative deepen sketches. We leveraged motherboards to address uncertainty through sketching and aesthetic intuition, refining options and ultimately selecting an optimal design. In the evaluation phase, we initially constructed a parametric 3D model with 20 parameters based on the optimal design, and invited 32 designers conducting participatory design experiments, getting 1024 human-designed schemes. Following this, we administered an online survey to assess the aesthetic qualities of a total of 1024 design schemes. Leveraging the collected score data (The first round of surveys engaged 279 participants, while the second round involved 73 participants), we trained an Artificial Neural Network (ANN) model to serve as an aesthetic evaluation score predictor for unknown parameter configurations. The machine could evaluate designs autonomously, thus selecting best design from 20,000 schemes generated randomly by machine. We utilized the parametric design converting sketching images to the numeric parameters, switching the qualitative ideation to the quantitative evaluation, thus achieving aesthetic evaluation and optimization. This study explores the relationship between human cognitive intuition and machine intelligence and how they can collaborate with each other.