( 1. 西北工業(yè)大學(xué) 機(jī)電學(xué)院， 西安 710072;
2. 西北工業(yè)大學(xué) 自動(dòng)化學(xué)院， 西安710072;
3. 西北工業(yè)大學(xué) 材料科學(xué)與工程學(xué)院， 西安 710072)
摘 要: 針對(duì)液-固擠壓復(fù)合材料管、 棒材成形時(shí)工藝參數(shù)難于選取、 試驗(yàn)工作量大的問(wèn)題， 在正交試驗(yàn)的基礎(chǔ)上， 結(jié)合有限元模擬數(shù)據(jù)， 構(gòu)建200組樣本集， 將其中的150組作為訓(xùn)練樣本用于網(wǎng)絡(luò)的訓(xùn)練學(xué)習(xí)， 其余的50組作為測(cè)試樣本用于驗(yàn)證網(wǎng)絡(luò)的精確性。 通過(guò)對(duì)補(bǔ)償模糊神經(jīng)網(wǎng)絡(luò)學(xué)習(xí)算法實(shí)現(xiàn)中的關(guān)鍵技術(shù)問(wèn)題的處理， 如輸入、 輸出變量模糊集的劃分、 模糊規(guī)則的提取、 學(xué)習(xí)速率的確定等， 基于模糊神經(jīng)網(wǎng)絡(luò)建立了液-固擠壓復(fù)合材料工藝系統(tǒng)模型， 得到了浸滲時(shí)間與其它關(guān)鍵參數(shù)之間的映射關(guān)系及模糊規(guī)則， 利用該模型， 對(duì)關(guān)鍵工藝參數(shù)進(jìn)行預(yù)測(cè)， 預(yù)測(cè)值與試驗(yàn)值吻合較好。 這為該工藝的實(shí)際應(yīng)用和過(guò)程控制奠定了基礎(chǔ)。
關(guān)鍵字: 模糊神經(jīng)網(wǎng)絡(luò); 復(fù)合材料; 液-固擠壓; 建模
（ 1. School of Mechatronic, Northwestern Polytechnical University, Xi'an 710072, China;
2. School of Automation, Northwestern Polytechnical University, Xi'an 710072, China;
3. School of Materials Science and Engineering,
Northwestern Polytechnical University,Xi'an 710072, China）
Abstract: Liquid-solid extrusion process, as a method of forming tubes, bars from liquid metal in a single process, is a kind of new metal forming technology, which was developed in recent years. But there exist some problems for forming the composite tubes or bars by this process, such as the difficulty of selecting process parameters and large quantity of the experiments required. In order to deal with these existing problems, on the base of the orthogonal experiments and FEA simulation, 200 groups of samples are constructed (150 groups are used to train the network, and 50 groups are used to verify the network), and the system model for liquid-solid extrusion is established by the compensatory neurofuzzy network (CNFN). Many key techniques in the realization of CNFN learning algorithms, such as the distribution of fuzzy sets for input and output variables, the determination of fuzzy rules and learning rate, are solved. By the established model, the relation among the infiltration time and other parameters can be mapped, and the key process parameters for extruding composite bars are forecasted. The forecasted and experimental results are well matched. So the present work builds a foundation for the reasonable choosing of the process parameters and practical application of the liquid-solid extrusion.
Key words: neurofuzzy networks; composites; liquid-solid extrusion; modeling
