神经网络在超音速等离子喷涂涂层摩擦学分析中的应用

发布时间：2018-03-05 06:07

  本文选题：超音速等离子喷涂　切入点：人工神经网络　出处：《江西理工大学》2012年硕士论文　论文类型：学位论文

【摘要】：摩擦磨损是造成零件失效的一个很重要的原因，所以零件耐磨性能的好坏在一定程度上决定了零件的使用寿命。研究证明，在零件表面制备一层具有润滑耐磨性能的涂层，可以显著地改善表面的磨损状况。在材料的摩擦学方面，存在着许多无法用明确的函数表达式来描述的非线性问题，而这些问题对了解材料的摩擦磨损性能有着很重要的作用，而人工神经网络很适合来处理这个问题，它可以完成n维空间矢量到m维空间矢量的映射。 本文采用超音速等离子喷涂技术在基体GCr15表面制备了KF-301/WS2复合涂层，利用MM-U5G屏显示材料端面高温摩擦磨损试验机对涂层进行摩擦磨损试验，然后分析其摩擦磨损性能。采用正交试验方法对影响涂层摩擦学性能的因素进行分析，确定较优的配方组合。以实验数据为基础，以温度、摩擦时间、润滑剂含量和表面微造型为输入量，，摩擦系数和磨损量为输出量，建立了一个4×7×2的三层BP神经网络，通过网络模型对样本数据进行训练学习，然后用训练好的网络对涂层进行摩擦磨损性能的预测分析。 研究结果表明：当温度在300℃~600℃时，磨损量和摩擦系数随着温度的不断升高而增大，但增长较缓慢，而当温度在600℃~750℃时，摩擦量和摩擦系数随着温度升高增长较快。在同一温度和同一WS2含量的情况下，不同微造型面的摩擦磨损性能从高到低依次是凹坑、菱形、平行、断纹。温度和表面微造型相同时，WS2含量为30%时的磨损性能要比WS2含量为20%时稍好一些。WS2含量为40%时，摩擦性能最差。当配方组合为润滑剂含量30%，表面微造型为凹坑时，涂层的摩擦磨损性能较好一点。通过三层BP神经网络的分析，预测结果和试验结果总体上拟合的比较好，预测结果所反映的规律和试验结果所反映的规律吻合，预测精度较高，因此所建神经网络模型可以对涂层摩擦性能进行预测分析。
[Abstract]:Friction and wear is a very important reason for the failure of the parts, so the wear resistance of the parts determines the service life of the parts to a certain extent. It is proved that a coating with lubricating and wear resistance is prepared on the surface of the parts. In tribology of materials, there are many nonlinear problems that cannot be described by explicit functional expressions, and these problems play an important role in understanding the friction and wear properties of materials. The artificial neural network is suitable to deal with this problem. It can map n-dimensional space vector to m-dimensional space vector. In this paper, KF-301/WS2 composite coatings were prepared on the surface of GCr15 substrate by supersonic plasma spraying technology. Friction and wear tests were carried out on the coatings by MM-U5G display material end surface high temperature friction and wear tester. Then the friction and wear properties of the coating were analyzed. The factors affecting the tribological properties of the coating were analyzed by orthogonal test method, and the optimum formula was determined. Based on the experimental data, temperature and friction time, A three-layer BP neural network of 4 脳 7 脳 2 is established, in which the lubricant content and the surface microform are input and the friction coefficient and wear quantity are output. The training and learning of the sample data are carried out by the network model. Then the friction and wear properties of the coatings are predicted and analyzed by the trained network. The results show that when the temperature is 300 鈩

本文编号：1569037