Convolutional Neural Networks and Applications on Civil Infrastructure

Abstract

Being a type of feed-forward Artificial Neural Networks (ANNs) with alternating subsampling and convolutional layers, Convolutional Neural Networks (CNNs) have been in widespread use in numerous Machine Learning tasks. Following the preliminary use of Deep 2D CNNs, 1D CNNs have been relatively recently developed and reached superior performance in various disciplines in a short time. Involving scalar multiplications and additions for 1D convolution, the compact and simple configuration has been the competitive edge of 1D CNNs. In this chapter, the authors are describing the evolution of Artificial Neural Networks (ANNs) into Convolutional Neural Networks. To portray the history of the evolution process, this chapter provides a comprehensive background and describes the way ANNs have progressed to the development of CNNs, covering the major steps and concepts pertinent to understand the way 1D CNNs operate. The chapter is organized into four sections. The first section focuses on the motivation behind the development of different Artificial Intelligence (AI) systems and subsets; the second section provides an overview of Artificial Neural Networks. The third section focuses on a particular type of ANNs called Multi-Layer Preceptors (MLPs); and finally, the fourth section discusses the training of 2D and 1D CNNs. Lastly, the applications of 1D CNNs on civil infrastructure are briefly discussed.