Fiber enhances polymer (FRP) the treatment of composite material is an issue of complex much physics field, involve conduct heat pass qualitative, thermochemistry photograph to change and height is nonlinear the viscoelasticity stress related to time develops. To reduce production risk and total manufacturing cost, now more keys are put on process imitate nowadays, in order to reduce blemish, is not to depend on technical know-how and relapsing experiment. This kind of number endows with the emulation setting 　 of 01 　 composite material component can the method uses global trade normally finite yuan (FE) simulation tool, wait like ANSYS of French ABAQUS, United States and Canadian COMPRO special and finite yuan of tool, include in afore-mentioned tools inside buy material library, evolve with showing those who machine function of the composite material in the process is complex. At present the industry of this kind of method is carried out rely on pair of spare partses that are heated by convection and tool to undertake costly three-dimensional and finite yuan of analysis. In predesign stage, need undertakes for many times designing iteration, with deciding spare parts ply eventually, fold the detail such as material of cycle of layer, solidify, mould and geometrical structure. To given large part, be like envelope of composite material wing, use 3D is finite yuan imitate undertakes the likelihood needs to count week of time analytic. To quicken process imitate, can use in predesign stage fall rank finite yuan replace 3D finite yuan. For example, learn an analysis to the thermalization of composite material, of small part basically spreading hot mechanism is to pass ply, be far from the brim and tool child structure. Accordingly, the good approximation that 1D FE analysis can answer as spare parts 3D undertakes, complex part can differentiate for distinctive area, in order to carry out analysis of many 1D FE, is not the analysis of complete 3D FE of whole spare parts. This accelerated the process imitate of predesign stage. However, this needs the balance of pair of speed and fidelity, if pursue,1 is shown.

The balance 　 between the fidelity that in machining a process, undertakes to composite material component thermochemistry analyses different solution and imitate time however, arrive from initial design detailed design, need finishs a large number of 3D finite yuan of imitate. Although use,can come true fall rank process simulative software, if use CoProducibility of Nvergent Manufacturing Technologies is evaluated - hot analysis (CPA-TA) , the imitate of component of large composite material still may need a few minutes to arrive one hour. Although as finite as complete 3D yuan of photograph comparing, this is a distinct advantage, but it is not quite fast still, cannot explore effectively whole optimize design limits. Composite material emulates 02 　 the new progress 　 of the design learns in view of the machine (Machine Learning, ML) the new progress with data drive method, a lot of branch of science and project had begun to realize ML for different application. A burgeoning application is to train to replace ML model quickly, in order to replace slow fast finite yuan of simulation tool. In this method, FE model is used at generating mass data automatically according to inputting the different hypothesis of parameter. Next, these data can be used at training different ML model, like nerve network (NN) , random forest model or gauss process are returned to (GPR) model. An acting model that passes proper training can duplicate well those who be used at training it is finite yuan of model, but have on imitate speed rise significantly. Close, the Navid Zobeiry that is in Washington university is professor, British the Anoush Poursartip of Colombian university is taught and come from CoIn the cooperative effort between the group of Nvergent Manufacturing Technologies, besides fall rank FE builds a model besides, still developed replace ML model to machine imitate in order to quicken composite material. In this research, developed replace nerve network model to replace fall rank finite yuan. These models use CoThe theory of the RAVEN software of Nvergent Manufacturing Technologies and problem of composite material solidify, the data that use FE imitate generates undertakes training to these models. These nerve network models can be used with finite the crucial function index during the input with yuan of identical imitate will forecast composite material to machine, include the temperature of tall spare parts during radiative solidify reacts. Average and character, use a typical computer workstation, use these ML models to be able to obtain the model that compare FE big 1000 to the speed gain of 10000 times. This makes the imitate speed of component of large composite material adjacent real time. What 03 　 machine learns emulation design is efficient models of these acting nerve network pass 　 nearly Co　 of Nvergent Manufacturing Technologies comes true in CPA-TA, showed their performance in the process imitate of envelope of wing of large composite material. Graph 2 in showed have envelope of representative 5×10 rice wing to design, its typical layer definition includes ply to be 40 go to 80, and the gasket in typical wing envelope, layer falls and other feature. The commercial software in use ABAQUS wraps COMPRO, the wing envelope that uses 3D FE to go up to manufacturing a tool on typical computer workstation undertakes process imitate, cost several hours. Notable is, use this kind of method to undertake flow optimizes need day or a few a few weeks time. In trying the 2nd times, wing envelope is differentiated to be 30 unique 1D geometry system. Use fall rank finite yuan, undertook imitate 9 times to whole wing envelope, use 3 temperature loop and ply of 3 kinds of cutting tool to undertake optimizing. These imitate that in CATIA 5 use CPA-TA undertakes are being used on the same table computer of 3D FE to cost about half hours. Demonstrate after in, the nerve network model that training passes in use CPA-TA undertook identical imitate, obtained first-rate result. With fall rank FE is similar, wing envelope cent is 30 area, used different mould ply and air temperature section plane to undertake imitate 9 times. However, these imitate are finished inside two seconds only. The comparison of these imitate methods (graph 2) ability mixes the imitate of close real time that proved ML method can achieve structure of complex composite material clearly to save time considerably. Notable is, the success of the technique that discusses here also can apply at other project application. (referenced origin: CW) starting fund former Dr. Xin