Research on product functional requirement prediction and automatic generation of design scheme based on deep learning

The study proposes a phased joint training strategy that integrates the collaborative optimization of functional requirement prediction and generative design. By employing cross-modal alignment, historical data is transformed into a unified semantic embedding space. A time-aware deep residual network (Time-aware ResCNN) is utilized to forecast future requirement trends and weight distributions. Subsequently, based on these predictions, a Conditional Progressive Generative Adversarial Network (CP-GAN) is adopted to generate diverse design solutions. The experiment is verified by the self-built cross-modal product data set ConsumerTech-2024. The results show that Time-Aware CNN completely surpasses the traditional model in F1-score, MAE and trend hit rate of demand prediction, and the complete model performs well in generating MMD values of design quality and manual scoring. Ablation experiments show that time convolution module (TC) and cross-modal consistency loss (CCL) have significant contributions to the model performance. The research results show that this method can effectively shorten the design cycle, improve the design efficiency and innovation, and the generated design scheme is more in line with the market demand and consumer preference, which provides new theoretical and methodological support for the product design field.