Surrogate-based Modeling And Optimization Of The Bleach Washing For Denim Jeans

Denim garments have received a wide acceptance in the world since Levi was granted a patent to produce work-wears made of denim fabrics and strengthened by copper rivets.In the evolution of denim garments,one of the important factors that has made denim garment go to the fashion stage is its unique washing process.The washing process relies heavily on the experience of the washers,and less young men are willing to become washers since poor working environment,so the industry has been suffering from an increasing labor cost year by year.On the other hand,washers only seek realizations of the water washing effect instead of optimal parameters which can achieve requirements and obtain minimum cost.Therefore,this paper proposes a method of using the approximate model to replace the experience of washers.Based on the proposed approximate model,this paper also establishes the cost optimization model and provides solving method.Because there are many kinds of washing techniques,this paper mainly selects three kinds of commonly used bleaching methods(hydrogen peroxide bleaching,sodium hypochlorite bleaching,enzyme bleaching)as the starting point.The approximate model of bleaching process can reflect the corresponding relationship between the washing process parameters and the product performances after washing.The accuracy of the approximate model is mainly influenced by the sampling method and the modeling method.To seek a better model,this paper adopts orthogonal sampling and Latin hypercube sampling respectively,and combining with polynomial response surface modeling method and the neural network model to construct four kinds of approximation model based on the data of bleaching experiments.This paper finds that the orthogonal sampling combined with quadratic response surface model can obtain better precision through the accuracy verification and comparison between models.The reason mainly lies in two aspects,firstly,because the limited samples,performance fluctuations and RBF neural network itself can lead to over fitting phenomenon,the quadratic response surface approximation model achieve better accuracy than RBF neural network.Secondly,the distribution of samples caused by the orthogonal sampling is better in a small sample size,so the results of the orthogonal sampling slightly better than using Latin hypercube sampling results.Based on the establishment of the approximate model,this paper also aims at optimizing the cost of bleaching with the constraints of the performance requirements and the input parameter ranges.The essence of the proposed optimization model is the nonlinear programming problem in mathematics,which is not easy to solve.However,the problem can be converted to integer nonlinear programming problem according to actual production practices,and can be solved by enumeration method.In this paper,we also make a case study to illustrate the application of the optimization model in actual production and prove the practicability of the model.