Research On Approximation Modeling And Knowledge Processing Methods For Product Scheme Design

The basic connotations of industry 4.0 are intelligence,greenness and humanization.Intelligence is the important foundation to realize greenness and humanization,and it is an important factor that affects efficiency and cost of industrial production.As one of the countries with complete industrial system in the world,improving intelligence is the fundamental way to realize the 2025 goal of Chinese manufacturing.Industrial product design is the premise of industrial product manufacture.With the structure and function of mechanical and electrical equipment becoming more and more complicated,design process is integrated,design knowledge is abundant,and design parameters are complicated,designers are difficult to obtain design results by single design process,and need to accomplish design task by repeated iterations,trials and optimization,so design efficiency is low and quality is difficult to be guaranteed.Therefore,in this paper,the approximation modeling methods for product scheme design with multi-type parameters are researched to support designers quickly to obtain product design scheme initial domain,knowledge processing method oriented various design knowledge is researched to support designers to obtain accurate product design scheme based on initial domain,and product scheme design support system is developed using the research findings,all of which are validated by the design process of large-scale power generation equipment.Production patterns of complex mechanical and electrical products are mostly single piece,small batch,and large sets,it is difficult to ensure that enough samples are obtained by multiple sampling because of samples being small and multi-type.For design case samples being sparse,variant parameters radial basis function response surface modeling method is proposed,by optimizing the parameters in radial basis function to accomplish modeling using a few sam ples.Experiments showed that RMSE accuracy is improved by more than 14.5% in the functions test,and the rotor shaft design case of hydro-generator proved that this method is able to realize the approximation modeling for the small samples design.For des ign case samples being missing,continuous function prediction model is proposed,by approximate supplementing the missing samples to accomplish modeling for design cases with missing samples.The method made the average relative error of test samples be reduced by more than 20% in the experimental data of NN3 chaotic time series,and the flow velocity coefficient selection of turbine spiral case also proved that the method is able to realize the approximate modeling for design cases with missing samples.Besides discrete parameters,many interval parameters also exist in complex product design.For traditional learning methods being difficult to realize the approximation modeling for production scheme including interval parameters,hybrid rules network model is proposed based on fuzzy mathematics theory in this paper.The model uses dynamic rule selection mechanism to select the optimal rules for the samples,and then uses the optimal rules for samples calculation to complete modeling.Experiments used Iris public data to verify the effect of different model parameters on results,and leave-one-out cross validation experiment proved that this model can achieve 96.67% accuracy.To further improve operation efficiency and stability,adaptive dynamic optimizing network structure method and adaptive hybrid rules network model are proposed.The validity and stability of the model was verified by Box-Jenkins gas furnace experiment,and hydro-generator electromagnetic winding design case proved that the model and met hod proposed in this paper can realize the approximation modeling for product design case including interval parameters.Graph case composed of continuous curves is difficult for rapid application in complex product scheme design.For both sample size afte r discretization and global error by traditional approximation modeling method being large,informative feature sample space establishing method based on nuclear pore is proposed.This method is used to obtain informative feature samples representing the f eature of continuous curves through filtering feature samples and judging importance of feature samples,so the approximate representation of graphical cases is achieved and the approximation model for graph case is established.Experiments showed that for Box-Jenkins gas furnace data,the model can maintain the similarity between the original curve and the training curve in the case of about 50% of the samples being reduced.Hydraulic turbine type selection design based on information feature samples showed that the proposed model can improve the efficiency of hydraulic turbine type selection design.For it being difficult for designers to quickly design complex product scheme,multi-type knowledge process method in product design is proposed.The proposed multi-type knowledge expression method based on Java-XML is used to achieve the knowledge modularization method oriented types.Design knowledge template is rapidly established based on work-flow model.Meta reason machine for knowledge module and two level collaborative reason mechanism for knowledge template are proposed,and knowledge module re-usability is improved.Reasoning buffer pool strategy is also proposed to improve the efficiency of multi-type knowledge reasoning.The multi-type knowledge processing method improves both product design knowledge modeling efficiency and engineering applicability.To improve the design efficiency of large-scale power generation equipment,large-scale power equipment design support system is developed,and the prop osed models and methods are also validated by the developed system.The researches in this paper are of theoretical significance for enriching product scheme design methods,and the example for the application of artificial intelligence technology in product design is provided.