The Optimal Design Of Plastic Texture Based On Artificial Neural Networks And Genetic Algorithm

In product design,material has gradually transformed from the physical basis of product function realization to the main medium which transfers the unique visual and tactile features in product emotion.As the most common material,plastic has high design flexibility and functional derivation.Therefore,through the intelligent calculation based on the quantitation of the users' needs expressed by the natural language,the plastic texture design is taken as an example in this research to output the optimal design solution of material texture which satisfying the users' needs.This method can fill a vacant position in research of the texture design solution generated by the natural language described needs and offer a scientific and effective guideline to the optimization design which puts the users' needs in center.Firstly,the words frequency analysis and clustering analysis are used to extracted the target images which describes users' affective needs.Through the simulation of the rendering software and the factor analysis,the design elements which make a difference on the visual effect of plastic texture and the value range of the elements are confirmed.After that,the value points of the elements are gained by the equidistance and the plastic material samples database is built through the orthogonal experiment.The quantitative data of all target images for every samples is obtained by the semantic difference survey,which can be used for training the artificial neural networks as the sample dataset.Secondly,combined with the trial and error method,the BP-ANN model is built to reveal the association between the plastic texture design elements and the target images.The network with the highest precision is chosen after 80 times training.It is verified by the validation set that the precision of the BP-ANN model is within expectation.The influence on the evaluation of target images made by the continuous change of design elements is analyzed and summarized.Finally,with the relative error between the value of images outputted by the BPANN model and the expected value controlling the direction of searching,the GA is applied to search the solution space of design elements to obtain the optimal value.The accuracy comparison shows that accuracy of GA improves sharply modified by the parallel selection method of the multi-objective.It is proven by validation data that this method can fit the target images evaluated by users well and output the value of the design elements whose error with actual value of the samples is less than 10%.It means the method is able to help and guide the designers to design the plastic texture under the users' needs.