面向AI生成的产品概念设计方案智能评估方法  

Intelligent evaluation method of product concept design scheme for AI generation

作　　者：王愫 刘月林[1] 孙利[1] WANG Su;LIU Yuelin;SUN Li(School of Arts and Design,Yanshan University,Qinhuangdao 066000,China)

机构地区：[1]燕山大学艺术与设计学院,河北秦皇岛066000

出　　处：《计算机集成制造系统》2025年第1期20-34,共15页Computer Integrated Manufacturing Systems

基　　金：国家社会科学基金艺术学资助项目(22BG125)。

摘　　要：面向批量生成的产品概念设计方案,为了实现在方案初筛阶段的高效且精准的智能化评估,提出一种基于改进的卷积神经网络的方案选择方法和评估模型。首先通过主成分法进行指标降维,筛选出产品概念设计方案具有代表性的指标,其次为具有更高的适用性,通过调查问卷建立结构方程模型,验证评估认知逻辑的合理性并得到评估指标的权重,作为数据集标注的依据。以头戴式耳机为研究案例,分别构建了带有方案感知价值标签的三分类和二分类数据集进行二分类对比实验,验证了方案图像分类效果与各评估指标的相关性。然后基于ResNet算法和卷积注意力机制对三分类数据集进行训练,获得方案图像智能评估模型,输出结果通过SHAP(Shapley Additive Explanations)算法进行可解释性分析,以助于设计师明确设计重点,为产品设计方案的初步筛选与设计迭代提供参考。将模型与其他经典卷积神经网络模型进行对比实验,结果表明产品概念设计方案评估模型的有效性和可行性。Facing the batch-generated product concept design,to provide efficient and accurate intelligent evaluation at the initial selecting stage of the concept,a concept selection method and evaluation model based on improved convolutional neural networks were proposed.The indicators were dimensionality reduction through principal component analysis to identify the representative indicators of the product concept design.To have higher applicability,the structural equation model was established with questionnaire for validating the reasonableness of the assessment of cognitive logic and obtaining the weights of the assessment indicators as a basis for dataset labeling.Taking headphones as a research case,three-classified and two-classified datasets with perceived value labels were constructed for two-classification comparison experiments respectively to verify the correlation between the product image classification effect and each assessment index.Based on ResNet algorithm and convolutional attention mechanism,the three-classification dataset was trained to obtain the intelligent evaluation model of concept design,and the output results were analyzed using the Shapley Additive Explanations(SHAP)algorithm for interpretability,which helped designers to clarify the design focus and provided a reference for the initial selecting and design iteration of product concept design.The model was compared with other classical neural network,and the results showed the effectiveness and feasibility of the product concept design evaluation model.

关 键 词：生成式人工智能 产品概念设计 结构方程模型 卷积神经网络 设计评估 

分 类 号：U462.1[机械工程—车辆工程] TP181[交通运输工程—载运工具运用工程]