A multi-bias recurrent neural network for modeling milling sensory data

Modeling high dimension sensory data is a key issue for Cyber-Physical Manufacturing Systems especially for milling process due to: (a) Sophisticated characteristics of input signals and (b) The complex procedure of processing sensory data. In this paper, we provide an End-to-End data modeling platform i.e., a multi-bias randomly connected recurrent neural network that makes use of recurrent structure and multi-bias to achieve efficient and accurate modeling performances. In order to tune the parameters of the proposed recurrent neural network (RNN), we apply a sampling method called Zoom-In-Zoom-Out (ZIZO) that helps RNN to quickly find a set of appropriate weights. We apply our technique to an empirical data set collected from NASA data repository and show that our method provides more precise and efficient results than existing methods.