Deep learning-based supervised and unsupervised neural networks for analysing the characteristics of powder composite preforms

ABSTRACT

This highlights the importance of deep learning in analysing the characteristics of composite preforms in the manufacturing of mechanical components in the industries. The raw data of Powder Metallurgy Lab are having highly non-linear, noisy and interrelated data. To process this kind of data, both shallow and deep neural network models of supervised learning have been considered for predicting the properties of composites. This work proves that the deep forward neural networks have good generality in recognizing even 100% of independent unseen data. The composites under the manufacturing process may have pores, which will reduce the strength and life time of the materials. Hence the sintering and extrusion of material processes are being followed to reduce the strength of pores. The presence of pores are examined by generating Scanning Electron Microscope (SEM) images for Cu–(5–20%)W composite preforms with a density of 94% without destroying the materials before and after the sintering and extrusion process. The pores are analyzed using unsupervised neural network model with Deep learning paradigm. These deep-learning based supervised and unsupervised models will guide the Lab Engineers to avoid the expensive experimentation and risky environment while preparing sintered composite preforms.

KEYWORDS:

• Deep neural network
• computational intelligence
• powder metallurgy
• composite preforms
• pore analysis
• sem images
• supervised and unsupervised learning

Disclosure statement

The authors would like to express their sincere thanks to the Management and the Principal, Mepco Schlenk Engineering College for showing constant encouragement to pursue this research work.

Additional information

Notes on contributors

Radha Pavanasam

Dr. P. Radha is working as a Professor in MCA Department of Mepco Schlenk Engineering College. Her qualifications are M.Sc., M.Phil. & Ph.D. in Computer science. She has 28 years of teaching experience. Her areas of interest are soft computing, IoT, Programming in C, C++, Java and open source tools. She has published 18 papers in the International Journals / Conferences. She has received sponsored research projects from AICTE and DRDO. One of her Journal Paper was Ranked 21st in the Science Direct Top 25 list of most downloaded articles in 2015 by  Science Direct Team, Elsevier Science Publishers.

Chandrasekaran G.

Dr. G. Chandrasekaran is a Senior Professor and Director of MCA Department of Mepco Schlenk Engineering College. His qualifications include M.Sc. (Applied Mathematics), M.S. (Software Systems), M.Tech. (Computer & IT), Ph.D. (Non Linear Systems). He has 38 years of experience in teaching and research. His areas of interest are Non linear systems, Software Engineering, E-commerce, Virtual Reality, Soft computing and Linguistic computing. He has published 80 papers in the International Journals / conferences. He has completed two research projects of AICTE.

N. Selvakumar

Dr. N. Selvakumar is working as a Senior Professor, Department of Mechanical Engineering, Mepco Schlenk Engineering College, Sivakasi. He has completed his Ph.D in the area of Powder Metallurgy. He is having rich teaching and industrial experience in Engineering Colleges and PSG Industrial Institute, Coimbatore, India. He has published more than 100 papers in reputed Journals, both International and National level. He is a recognised research supervisor for Anna University and till date produced 14 Ph.D graduates and pursuing 5 more Ph.D research Scholars. He has developed the Centre for Nano Science and Technology at Mepco. He is also a Principal investigator for a sponsored research project funded by ISRO, AICTE & DRDO. He is authored for two text book in Metallurgy published by Scitech, Chennai, India.