基于机器学习确定膜基材料弹性模量的仪器化压痕方法

Determining the elastic modulus of film/substrate materials from instrumented indentation testing based on machine learning

仪器化压痕测试方法在确定材料的力学性能方面有着广泛的应用.由于基底效应,采用该方法确定膜基材料各层的力学性能变得更为复杂.本文采用有限元仿真结合经典多输出多层感知器(MLP)神经网络的方法,建立了膜基材料力学性能参数(薄膜弹性模量、基底弹性模量)与复合模量和无量纲化最大压入深度(最大压入深度/薄膜厚度)之间的关系,并发展了基于机器学习确定膜基材料弹性模量的仪器化压痕方法.对比深度学习的预测和有限元的仿真结果,发现经MLP神经网络训练的硬膜软基底材料和软膜硬基底材料的预测值与仿真结果的吻合度较好.开展硬膜软基底材料Ni/304不锈钢以及软膜硬基底材料Cu/304不锈钢的压痕试验对MLP神经网络进行验证,结果表明, MLP神经网络预测的各层弹性模量与试验中获得的结果较为接近.本文结果可为评价膜基材料各层性能提供可供选择的测试方法.

Instrumented indentation testing has been widely used to determine the mechanical properties of materials. However, due to the substrate effect, the mechanical properties of each layer of film/substrate material becomes complicated to determine using this method. Through the finite element simulation, combined with a classic neural network method of multi-output multilayer perceptron(MLP), the relationship between the film/substrate material parameters(elastic modulus of film and substrate), normalized indentation depth(indentation depth/film thickness), and composite modulus is established. This established relationship is used to develop an indentation method to determine the elastic modulus of film/substrate. Prediction results through deep learning and the finite element simulation results are compared. The comparison results indicate that the predicted values of hard film/soft and soft film/hard substrate materials obtained from MLP are in good agreement with the simulation results. Indentation tests of Ni/304 and Cu/304 stainless steel were conducted to verify the trained neural network. The results indicate that elastic modulus of each layer predicted using the MLP is close to those obtained in the test. The results of this study can provide alternative research methods for evaluating the properties of film/substrate materials.