【作者】 刘海定；

【导师】 汤爱涛； 潘复生；

【作者基本信息】 重庆大学 ， 材料物理与化学， 2005， 硕士

【摘要】 近年来,镁及镁合金的发展和应用非常迅速,尤其是镁合金制备和处理工艺的进展大大推动了其在汽车、电子通讯及航天航空工业的应用。但是镁合金仍存在许多问题,比如镁合金熔炼难、强度低、高温性能不好、耐腐蚀性能较差以及室温塑性不佳等等,这些问题严重制约着镁合金的大规律工业化应用。本文广泛收集了镁合金化学成分、处理工艺及力学性能等方面数据,并对数据进行科学的整理和分析,采用神经网络技术建立镁合金化学成分、处理工艺及力学性能之间的定量关系模型,模型的输入包括镁合金的主要合金元素含量,铸造或变形处理工艺,热处理的温度和时间,以及测试温度;模型的输出为三个力学性能指标,即抗拉强度、屈服强度和延伸率。最后利用模型对镁合金化学成分、处理工艺及力学性能之间的关系规律进行较为系统深入的研究。论文首先从数据的预处理方式、中间隐层结构、学习算法和学习参数四个方面探索镁合金神经网络模型的最佳选择。结果表明,上述四个因素均会影响模型的收敛,其中学习算法的影响最为显著,而学习参数的影响较小。采用LM 算法能得到比其余5 种算法更佳的效果。镁合金抗拉强度、屈服强度和延伸率的数据预处理方式分别为[0.1,0.8]、[0.01,0.99]和[0.01,0.99],中间隐层结构为单隐层,且单元数分别为10、11 和12 时,能取得最佳效果。论文其次应用最佳模型预测了AZ61B 及Mg-Zn-Zr-Y 变形镁合金的室温力学性能。预测结果与实验值大多吻合得很好,其中抗拉强度和屈服强度的效果尤佳,抗拉强度的最大相对误差不超过6.5%,屈服强度的不超过16%,但延伸率的效果较差,特别是对合金的突然脆化反应迟钝。此外,还对Mg-Al-Zn 合金的铸态室温力学性能进行预测,预测结果在可接受范围之内,预测的变化趋势可从显微组织分析得到印证。论文最后一部分重点对模型的应用进行了深入研究,主要有以下三个方面: ①研究了不同Zn、Zr 和Y 含量对Mg-Zn-Zr-Y 变形合金力学性能的影响,结果表明,Zn 含量的增加有利于合金抗拉强度、屈服强度的提高,但使合金延伸率降低,该变化趋势跟文献结果研究基本一致;而Zr 含量不宜过高,与Zr 含量为0.15%、0.35%和0.55%的合金相比,Zr 含量为0.75%的合金有更低的抗拉强度、屈服强度及延伸率。②研究了不同Al、Zn 含量对Mg-Al-Zn-Mn 铸态室温力学性能的影响。结果表明,随Al 含量的增加,合金的抗拉强度先增加后减小,当Al 含量为3.0%左右达到最大值,之后抗拉强度缓慢降低,延伸率表现出大致相同的规律,但是屈服更多还原

【Abstract】 Advances in manufactueing and processing technologies in recent years have stimulated renewed interest in magnesium alloys for applications in the automotive, communications and aerospace industries. However, there are some problems, such as prone to burning or oxidation while smelting, low strength, less resistance of creep at elevated temperature, poor corrosion resistance performance, and low plastic formability at room temperature, etc, which are major obstacles to usage of much bigger industry scale. In the present work, the available data on mechanical properties of magnesium alloys were collected from domestic and foreign literature, then data analysis was carried out. Meanwhile, a model was developed for the analysis and prediction of the correlation between the mechanical properties of magnesium alloys and compositon, processing and working condition by using artificial neural network (ANN) based on data. The input parameters of neural network (NN) model are alloy composition, cast or wrought processing parameters, heat treatment parameters and work (test) temperature. The output parameters of the model are three mechanical properties namely ultimate tensile strength (UTS), yield strength (YS) and elongation (ELO). Finally, the effect of alloying elements on mechanical properties of magnesium alloys was investigated and discussed. To find the optimum structure of the model, different options were investigated, including preprocessing method of data, hidden layer of NN model, algorithm and its main raining parameter. The results show that algorithm has the most remarkable effect, and the training parameters have the less sight effect. The Levenberg-Marquardt (LM) algorithm got the best performance than others algorithms. The others three factors mension above have some effects on model. The best preprocessing methods of UTS, YS and ELO model are [0.1, 0.8], [0.01, 0.99] and [0.01, 0.99], and best hidden layers are single hidden layer with 10, 11 and 12 neurons. The model was used to predict the mechanical properties of AZ61B and Mg-Zn-Zr-Y wrought alloys. The results show that the predicted mechanical properties coincided with the experimental data quite well, especially the performance of UTS and YS model much better. The relative errors between predicted ultimate tensile strength and experimental data are smaller than 6.5%, and these of YS model are not bigger than 16%. But the accuracy of the ELO model is not better than that of UTS and YS models. Especially the ELO model has great difficulty to predict the properties break of magnesium alloys, for an instance, the Mg-8.3Zn-0.57Zr-1.4Y alloy becomes brittleness after T6 heat treatment (500℃/3hours+170℃/24hours), but the model can not predict this. Moreover, the mechanical properties of Mg-Al-Zn as-cast alloys was predicted using the optimum model. Available results were obtained and were verificated by microstructure analysis. The model was also used to predict effects of alloying elements on the mechanical properties of Mg-Zn-Zr-Y alloys, Mg-Al-Zn-Mn alloys and Mg-Al-Si alloys. The results are as follows: (i) Zinc can improve the ultimate tensile strength and yield strength of Mg-Zn-Zr-Y wrought alloys at room temperature, and reduce their elongation. These predicting results very close to that in literature. The high content of zirconium is unfavourable, compare to the alloys with Zr content of 0.15%, 0.35% and 0.55%, the alloys with 0.75% Zr have rather lower tensile strenght, yield strenght and elongation. (ii) With the increase of aluminium content, the ultimate tensile strength of as-cast Mg-Al-Zn-Mn as-cast alloys at room temperature went up firstly and then went down. While Al content is 3.0%, the maximum tensile strength was obtained. After that, it decreased slowly. The yield strength has the same results, but elongation has not. And zinc has sligh effect on the mechanical properties. (iii) With silicon content increase, ultimate tensile strength of as-cast Mg-Al-Si alloys at room temperature decrease, and elongation have the same result, but yield strength have opposing result. The correlation between alloy composition, processing parameters and final properties of magnesium alloys is of importance and complexity. In the present work, modeling of them by using ANN has good performance, which indicate that the method is feasible and effective. However, some problems exist. For examples: (i) ANN has poor capability to get the best result which exceed its range; (ii) experimental data of magnesium alloys are still limit compared to the data in aluminium alloys and steels. Some other methods have been suggested to resolve to the first problem. As to the second prolem, simple model was developed to predit the mechanical properties of Mg-Al-Zn as-cast alloys. Compare to the results of unsimple model, better preforance was achieved. In addition, grain size of Mg-Al-Zn as-cast alloys was predicted by using ANN.The results indicate that with the increase of aluminium content, the grain size of as-cast Mg-Al-Zn becomes more finer that can be verificated by the microstructure analysis更多还原