Study of machine learning based algorithms for active noise control of machinary noise

Abstract

Machine learning is becoming a part of every field. After the discovery of deep learning, its application in various fields are increasing day by day. This paper presents a study of three machine learning based active noise control algorithms to reduce machinary noise. Three algorithms studied in this work are: filtered-x backpropagation neural network (FxBPNN), radial basis function (RBF) and deep recurrent neural network (DRNN). Experimentally recorded noises of band saw, CNC and compressor are used in the simulation study. Two cases of primary path are considered: a) linear path and b) nonlinear path, along with a linear secondary path. Performance of the three algorithms are evaluated based on steady state residual noise.