Effect of Activation Function and Post Synaptic Potential on Response of Artificial Neural Network to Predict Frictional Resistance of Aluminium Alloy Sheets

2017 4th International Conference on Advanced Materials, Mechanics and Structural Engineering

材料科学;机械制造

Trzepiecinski, T.^1 ; Lemu, H.G.^2

Rzeszow University of Technology, Department of Materials Forming and Processing, al. Powst. Warszawy 12, Rzeszów

35-959, Poland^1

University of Stavanger, Department of Mechanical and Structural Engineering, Stavanger

N-4036, Norway^2

关键词: Aluminium-alloy sheets;    Coefficient of frictions;    Frictional resistance;    Mechanical parameters;    Neural network modelling;    Post-synaptic potentials;    Technological factors;    Tribological conditions;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/269/1/012041/pdf
DOI  :  10.1088/1757-899X/269/1/012041

学科分类：材料科学（综合）

来源: IOP

【 摘 要 】

Many technological factors affect the friction phenomenon in sheet metal forming process. As a result, the determination of the analytical model describing the frictional resistance is very difficult. In this paper, a friction model was built based on the experimental results of strip drawing tests. Friction tests were carried out in order to determine the effect of surface and tool roughness parameters, the pressure force and mechanical parameters of the sheets on the value of coefficient of friction. The strip drawing friction tests were conducted on aluminium alloy sheets: AA5251-H14, AA5754-H14, AA5754-H18, AA5754-H24. The surface topography of the sheets was measured using Taylor Hobson Surtronic 3+ instrument. In order to describe complex relations between friction and factors influencing tribological conditions of sheet metal forming, the multilayer artificial network was built in Statistica Neural Network program. The effect of activation function and post synaptic potential function on the sensitivity of multilayer neural network to predict the friction coefficient value is presented. It has been found that the difference in the prediction of error of neural network for different approaches can reach 400%. So, the proper selection of activation and post synaptic potential functions is crucial in neural network modelling.