Data-driven Product Appearance Image Design Technique And Its Application

With the homogenization of product function and performance,design technique for product appearance quality has been increasingly valued by academia and industry,and has become an effective technical approach to meet the ever-increasing diverse and individual needs of consumers.The appearance quality attributes of a product can be characterized or described by product appearance image(PAI).Based on the quantitative data of product appearance and consumers' emotional needs,data-driven PAI design achieves the innovation of product appearance through computer-aided design technologies and intelligent algorithms.It can help companies quickly respond to consumers' emotional needs,and effectively improve the design quality and product competitiveness.However,there are some problems in the existing research,such as the limitation of the quantization technique of product appearance,the inaccuracy and comprehensiveness of the positioning and prediction of PAI,and the lack of systematic PAI design technique.To this end,by combining theory with simulation analysis,this paper comprehensively uses Kansei engineering theory,computer aided design technology and intelligent algorithm to conduct an in-depth and systematic study on various aspects of data-driven PAI design,such as product appearance quantification,PAI positioning and prediction,image-oriented product appearance generation,optimization and decision-making,and proposes a data-driven PAI design technique.The validity of the research results is verified through the application case of car appearance image design.The specific research content and results are as follows:(1)Mathematical model of product appearance quantification.First of all,based on the ellipse Fourier technique that has the advantages of feature extraction and image recognition matching for two-dimensional(2D)product form and principal component analysis,a mathematical model for the quantization of 2D product form is established,which solves the problem of the accuracy and applicability of image-oriented quantification of 2D product form.Secondly,the combination of spherical harmonic,triangular mesh model and auto-encoder,which have theadvantages of three-dimensional(3D)product form-oriented feature extraction and image recognition matching,is used to construct a mathematical model for the quantization of 3D product form,which solves the problem of the feasibility and applicability of image-oriented quantification of 3D product form.Thirdly,combining color theory and product coloring principle,a mathematical model for the quantization of product color based on the color model and coloring area division is constructed,which provides support for the realization of quantitative data acquisition and data fusion of different appearance elements.Finally,the mathematical model for the quantization of product appearance is formed by combining the mathematical models of 3D product form and color,which realizes a complete quantitative description and feature extraction of the appearance of complex products.(2)Positioning of PAI.In view of the lack of accuracy and comprehensiveness in the current PAI positioning,the quantitative data of product form,color and appearance is applied to the positioning,and the multi-image positioning methods of product form,color and appearance are formed.The methods study and determine the selection principle of sample and image vocabulary,preprocessing and image measurement technique.Based on correlation analysis,the comprehensive image evaluation models of product form,color and appearance are established respectively,and the positioning results are obtained using these models and related statistical analysis techniques.The results retain more comprehensive image information,reflect the difference in importance between different key images,and help to more scientifically and reasonably determine the multi-image targets and evaluation criteria for product form,color and appearance design.(3)Prediction of PAI.Aiming at the shortcomings of current PAI prediction techniques,the multi-image prediction models of product form,color and appearance are proposed respectively.First of all,according to the characteristics of form data obtained by harmonic technique,the concept of key principal component and its identification method are proposed.Using multiple linear regression analysis,Akaike information criterion(AIC)and Bayesian information criterion(BIC),a product form image prediction model based on key principal components is constructed and the model effect is verified.Secondly,according to the data characteristics of productcolor,the concept and identification method of key color variable are proposed,and a product color image prediction model based on key color variables and its effect verification model are further established.Finally,to ensure the image prediction accuracy of the fusion data of 3D form and color for product appearance,a genetic algorithm and back propagation neural network(GABP)-based multi-image prediction model of product appearance and its effect evaluation method are constructed.Each model fully combines the quantitative data characteristics of product appearance for data modeling,which makes up for the shortcomings of the existing prediction technique and improves the prediction accuracy and effect of product image.(4)Image-oriented generation,optimization and decision-making of product appearance.To form a technique system for PAI design,firstly,image-oriented generation and evaluation techniques of 2D form,3D form and color are proposed,including the generation and evaluation technique of 2D form based on key principal components and the adjustment of related form features,the generation and evaluation technique of 3D form based on conditional variational auto-encoder(CVAE)and its generation effect evaluation index,and the generation and evaluation technique of color based on key color variables and color multi-image optimization model.Secondly,to realize the intelligent design of the whole process of PAI design,an improved multi-objective optimization algorithm,improved strength Pareto evolutionary algorithm 2(ISPEA2),with better comprehensive performance is proposed and a more objective and accurate multi-attribute decision-making method,entropy weight and technique for order preference by similarity to ideal solution(entropy-TOPSIS),is selected.These methods ensure the innovation and image conformity of the multi-objective optimization of appearance schemes and the accuracy and rationality of the decision on optimal appearance scheme.Finally,an integrated technique system for complete PAI design is formed.(5)Applied research.The theoretical research results are applied to the engineering practice of car appearance design.The process of the specific implementation of data-driven PAI design is explained in depth,and the feasibility,effectiveness and application value of the theoretical research results are verified.The paper takes 2D form,3D form,color,and appearance combining 3D form and color as the research objects.From the aspects of product appearance quantification,PAI positioning,PAI prediction,image-oriented product appearance generation,optimization and decision-making,it systematically and innovatively solves the related problems that restrict product appearance quality design,forms a data-driven PAI design technique,and provides a scientific theoretical basis and analysis method for effectively carrying out product appearance design research and development and engineering practice.